JPH07152396A - Voice synthesizer - Google Patents

Abstract

PURPOSE:To provide a voice synthesizer with high quality capable of flexibly dealing with various voice characteristics, etc. CONSTITUTION:This device is provided with a voice sound source part 38, a series type formant synthesis part 32, a consonant generation part 39 provided with a consonant waveform storage part 33 and a synthesis part 35, and a vowel and a nasal are formed by the voice sound source part 38 and the series type formant synthesis part 32, and the consonant such as a plosive and a fricative, etc., is generated from a waveform stored in the consonant waveform storage part 33, and they are synthesized by the synthesis part 35. Both waveforms are phase synchronized by a pitch synchronizing signal generation part 24. Related to the vowel, flexible and various tone quality and intonation are added by a formant synthesis system, and related to the consonant, a high quality voice unrealisable usual formant synthesis system is provided by the system using a waveform element piece. Since the storage as the waveform element piece is limited to the consonant with a shorter duration time, it is realized by a storage with smaller capacity.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech synthesizer for converting arbitrary text into speech.

[0002]

2. Description of the Related Art There are roughly two types of speech synthesis methods for converting arbitrary text into speech. One is a synthesizing method that understands the vocalization mechanism of the voice, that is, the movements of the vocal cords, the mouth, and the throat, and controls the electric circuit by using the knowledge as a rule.
The other method is a method that does not require much knowledge of speech, and prepares a large number of speech segments and connects the appropriate segments according to the input. The former is well known, for example, a combination of a formant synthesis method and a formant control rule. FIG. 4 shows a configuration example of a combination of the formant synthesis method and the formant control rule. In the figure, a formant synthesizer control rule storage unit 9 stores a plurality of rules for controlling the formant synthesizer, and a formant synthesizer control coefficient generation unit 8 is a formant synthesizer based on the control rule. , A formant synthesizer 10 actually synthesizes a voice, a voiced sound source section 1 simulates a vocal cord vibration, and a serial formant synthesizer 2 forms formant resonators in series. The part that connects and synthesizes voiced sounds such as vowels and nasal sounds, the unvoiced sound source part 6 is a turbulent noise source necessary for synthesis of fricative sounds and plosive sounds,
The parallel formant synthesizer 7 has resonators connected in parallel and synthesizes unvoiced consonant parts such as fricatives and plosives. Synthesis part 5
Is a part for synthesizing the output of the serial formant synthesizer 2 and the output of the parallel formant synthesizer 7 and outputting a synthesized sound.

When information about phonetic symbols, accent positions, and intonation necessary for speech synthesis is input to the formant synthesizer control coefficient generator 8, the formant synthesizer control coefficient generator 8 stores the formant synthesizer control rule. The unit 9 refers to a necessary rule and outputs the formant synthesizer control coefficient to the formant synthesizer 10. The inside of the formant synthesizer 10 operates as follows. The voiced sound source unit 1 simulates a pulse-shaped sound source waveform generated in a human vocal cord when synthesizing a voiced sound such as a vowel. This pulse-shaped signal is input to the serial formant synthesizer 2, and the serial formant synthesizer 2 provides the sound source waveform with an appropriate feature as a vowel or a nasal sound by using a plurality of formant resonators and outputs it to the synthesizer 5. On the other hand, the unvoiced sound source section 6 sends a noise-like waveform, which is a sound source of fricative or plosive sound, to the parallel formant synthesis section 7, and the parallel formant synthesis section 7 uses a plurality of resonators to obtain the frequency characteristics required for each consonant. Are formed instantaneously and output to the combining unit 5. The synthesizer 5 synthesizes the vowels and nasal sounds of the serial formant synthesizer 2 and the consonants of the parallel formant synthesizer 7 and outputs them as synthesized speech.

Next, another method of using a waveform segment (speech segment) as a conventional example will be described. FIG. 5 is a block diagram of this system. The waveform segment selection unit 11 selects from the waveform segment database storage unit 12 a waveform segment necessary for synthesis from the phonetic symbol string and the accent information which are input. In this case, the waveform segment is usually compressed and stored in a coefficient such as a linear prediction coefficient. The selected plurality of waveform segments are connected by the segment connection synthesis unit 13 and synthesized into a speech waveform at an appropriate fundamental frequency.

[0005]

By the way, the inventors of the present invention have studied the above-mentioned two methods and found that the two methods have different characteristics as described below.

That is, the advantage of the former method is that since all the sounds are created according to rules, it is highly flexible and can synthesize voices of various tones and intonations. The disadvantage is that the synthesis rules have not been sufficiently studied, especially for voices with complex vocalization mechanisms such as consonants.
It is difficult to generate natural consonants.

Further, the advantage of the latter method is that since the waveform segment is basically cut out from the natural speech which is the model, the synthesis quality is extremely high if the segment can be smoothly connected. On the other hand, the disadvantage of this method is that it requires a large-capacity storage device to store the waveform elements, and thus the cost is high. Also,
There is a problem that only the voice quality of model speech can be synthesized, and it lacks flexibility.

In summary, in the case of a system in which all sounds are made up by rules, it is possible to synthesize voices of various sound quality and intonation with flexibility, but the synthesis rule is still clear for voices with a complicated voicing mechanism such as consonants. Since it is not done, it is difficult to synthesize. On the other hand, in the case of the method using the waveform segment, the synthesis quality is extremely high, but there is a problem that a large-capacity storage device is required for storing the waveform segment, and that only the voice quality of the model voice can be synthesized, resulting in lack of flexibility. There's a problem.

An object of the present invention is to provide a speech synthesizing apparatus having high synthetic quality, which is highly flexible in sound quality and can be significantly reduced in storage capacity as compared with the conventional method using waveform segments. .

[0010]

According to a first aspect of the present invention, a voiced sound source section for outputting a voiced sound source signal, and a plurality of series-formed formant resonances to which a voiced sound source signal from the voiced sound source section is input. Type formant synthesis section for synthesizing a predetermined sound such as a vowel, a waveform storage section for storing a waveform of a predetermined sound such as a plurality of consonants, and a waveform for reading out a required waveform from the waveform storage section A speech synthesizer comprising: a reading unit; and a waveform combining unit that outputs the synthesized speech by superposing or switching the output from the serial formant synthesis unit and the waveform read by the waveform reading unit.

According to a third aspect of the present invention, sound generating means for generating a voice signal based on the characteristic parameters extracted from the voice, and waveform element storage means for storing the waveform element cut out from the voice, A waveform segment feature amount storage unit that stores a predetermined feature amount for the stored waveform segment, and a voice signal generated by the sound generation unit based on the stored feature amount of the waveform segment and the voice signal. It is a speech synthesizing device provided with a control means for synthesizing a waveform segment signal obtained from a waveform segment storage means.

According to a seventeenth aspect of the present invention, a voiced sound source waveform generating means for generating a voiced sound, a serial formant synthesizer, a consonant sound waveform generating means for generating a consonant, a waveform connecting means for connecting a waveform, and a pitch synchronization signal generating means. The pitch synchronization signal generating means outputs a pitch synchronization signal corresponding to a desired pitch period, and both the voiced sound source waveform generating means and the consonant sound waveform generating means generate a waveform of a phase synchronized with the pitch synchronization signal. And the serial formant synthesizer changes the frequency characteristic with a transfer function simulating the vocal tract characteristic in the output waveform of the voiced sound source waveform generator, and the waveform connecting means outputs the output of the serial formant synthesizer. A voice synthesizer for generating a voice waveform by connecting or mixing the waveform and the output waveform of the consonant sound waveform generating means.

[0013]

According to the present invention of claim 1, the voiced sound source section outputs the voiced sound source signal, and the serial formant synthesis section receives the voiced sound source signal from the voiced sound source section as an input, and a plurality of formant resonances connected in series. And a predetermined sound such as a vowel is synthesized, a waveform storage unit stores a waveform of a predetermined sound such as a plurality of consonants, and a waveform reading unit reads a required waveform from the waveform storage unit. In addition, the waveform combination unit superimposes or switches the output from the serial formant synthesis unit and the waveform read by the waveform reading unit, and outputs the synthesized voice.

According to the present invention of claim 3, the sound generating means generates a voice signal based on the characteristic parameter extracted from the voice, and the waveform element storage means stores the waveform element cut out from the voice. Then, the waveform segment feature amount storage means stores a predetermined feature amount for the stored waveform segment, the control means, based on the stored feature amount of the waveform segment,
The voice signal generated by the sound generation unit and the waveform segment signal obtained from the waveform segment storage unit are combined.

In the seventeenth aspect of the present invention, the pitch synchronizing signal generating means outputs a pitch synchronizing signal corresponding to a desired pitch period, and both the voiced sound source waveform generating means and the consonant sound waveform generating means are synchronized with the pitch synchronizing signal. A phase form waveform is generated, the serial formant synthesizer changes the frequency characteristic with a transfer function simulating the vocal tract characteristic of the output waveform of the voiced sound source waveform generator, and the waveform connecting means outputs the serial formant synthesizer. A voice waveform is generated by connecting or mixing the waveform and the output waveform of the consonant waveform generation means.

[0016]

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

FIG. 1 shows the configuration of a speech synthesizer according to the first embodiment of the present invention. In the examples below,
The explanation will mainly be in Japanese, but it can be applied to other languages such as English as long as there is no problem. In the figure, the voiced sound source section 1 is a section that simulates vibration of the vocal cords and generates a sound source signal. The serial formant synthesis unit 2 is a unit that synthesizes a voiced sound such as a vowel. The consonant waveform storage unit 3 stores a waveform segment of a consonant cut out from a natural voice, the consonant waveform readout unit 4 selects and extracts a required waveform segment, and the synthesis unit 5 a serial formant synthesis unit 2 Is combined with the output of the consonant sound waveform reading unit 4 and is output as a synthesized voice.

The operation of the speech synthesizer of this embodiment having the above-mentioned structure will be described below.

As described in the conventional example, the formant synthesizer control coefficient is first given to the present synthesizer. The voiced sound source section 1 generates a desired sound source signal from the information on the fundamental frequency in the coefficient for controlling the formant synthesizer, the information on the amplitude information of the sound source, and the like, and the serial formant synthesizer 2
To enter. No sound source signal is output in the consonant section or the unvoiced section. The series formant synthesizer 2 determines the characteristics of the resonators arranged in series from the formant frequency information in the formant synthesizer control coefficient, the information about the bandwidth of the formant resonance peak, and the like, and determines the vowel sound from the sound source signal. Etc. to a voice signal. The output of the serial formant synthesis unit 2 is sent to the synthesis unit 5. On the other hand, the consonant sound waveform reading unit 4 confirms whether or not the phoneme exists in the consonant sound wave storage unit 3 from the information about the phoneme in the coefficient for controlling the formant synthesizer, and if it exists, the consonant is detected. It is taken out from the waveform storage unit 3 and sent to the synthesis unit 5. For example, as shown in FIG. 2, when the phoneme to be synthesized is “k” and the following vowel is “a”, the consonant sound waveform reading unit 4 stores the consonant sound “k” in the consonant sound waveform storage unit 3. Search for the waveform element cut out from ". The synthesizing unit 5 synthesizes the vowel signal from the serial formant synthesizing unit 2 and the consonant signal from the consonant sound waveform reading unit 4 by addition processing or superposition processing. By configuring in this way, it is possible to provide various sound quality and intonation for vowels by the formant synthesis method, and for consonants, high quality speech that cannot be achieved by the formant synthesis method by using the waveform segment is provided. it can. Since storage as waveform segments is limited to consonants with short duration, it can be realized with a small-capacity storage device.

Next, with reference to FIG. 3, the second embodiment of the present invention is capable of retaining the features of the above-mentioned method, reducing the types of waveform elements, and reducing the required storage capacity. The speech synthesizer in will be described.

In the figure, the unvoiced sound source section 6 is a consonant sound source, and the parallel formant synthesis section 7 is a burst sound or a fricative sound produced by a plurality of resonators connected in parallel with the signal from the unvoiced sound source section 6. This is the part that synthesizes consonants such as. Other means are the same as those in the first embodiment.

The operation of the speech synthesizer of the present embodiment constructed as above will be described below.

Similar to the first embodiment, first, the formant synthesizer control coefficient is given to the present synthesizer, and the voiced sound source section 1 and the serial formant synthesizer 2 convert it to a vowel signal and synthesizer 5 is added. Sent to. Further, among the consonants, those which can achieve sufficiently high quality by the formant synthesis method are handled by the unvoiced sound source section 6 and the parallel type formant synthesis section 7. That is, the unvoiced sound source section 6 adjusts the amplitude, timing, etc. of the noisy signal based on the information about the unvoiced sound source in the given formant synthesizer control coefficient, and sends it to the parallel formant synthesis section 7. In the parallel type formant synthesis unit 7, the noise signals are converted into desired consonant signals by the resonators arranged in parallel based on the information regarding the frequency characteristics of the consonants to be synthesized, and are passed to the synthesis unit 5. The consonant sound waveform reading unit 4 and the consonant sound waveform storage unit 3 are similar to those in the first embodiment.
A consonant which is not handled by the parallel formant synthesizer 7 is searched from the waveform segment database and sent to the synthesizer 5. Similar to the first embodiment, the synthesizer 5 uses the vowel signal from the serial formant synthesizer 2 and the consonant sound waveform read-out unit 4.
Or the consonant signal from the parallel formant synthesizer 7 is synthesized by addition processing or superposition processing. With this configuration, for consonants that can achieve sufficiently high quality with the formant synthesis method, the unvoiced sound source part and the parallel formant synthesis part can be used, and the storage capacity required for storing waveform segments can be reduced. Will be possible.

Further, by driving the parallel formant synthesizer 7 and the consonant sound waveform readout unit 9 simultaneously, for example,
In a noisy environment or the like, the intelligibility can be increased more than natural speech by, for example, further emphasizing the ruptured portion of a certain waveform segment with the signal from the parallel formant synthesis unit 7.

Next, another embodiment of the present invention will be described.

The consonant sound waveform storage unit stores a consonant portion cut out from a natural voice waveform. In the case of unvoiced consonants, consonant parts such as bursts and friction parts can be separated from voiced sound parts, i.e., parts after vocal cord vibration has started. It can be used for pitch synthesis. However, voiced consonants cannot separate the consonant part from the voiced part, so the waveform after the vocal cord vibration has started must be included in the segment.

In general, cues for the perception of consonants are also included in the following vowels. Therefore, the sound quality can be improved by including the beginning part of the following vowel in the consonant waveform segment.

Therefore, it is necessary to connect the consonant sound wave segment and the series formant synthesized waveform at the succeeding vowel portion. At this time, for example, if the series type formant composite waveform is instantaneously switched in the middle of the consonant waveform segment, waveform discontinuity occurs and impulsive noise occurs.

A method of performing smooth superposition with a predetermined section width can be considered. That is, the consonant sound wave segment is smoothly attenuated, and the series formant composite waveform is smoothly raised. If the first and second pitch periods of the succeeding vowel portion are included in the consonant sound wave segment and overlapped with a section width of about one pitch period, the consonant sound wave segment can be used without considering the pitch.

However, even if connection is made by the above method, phase discontinuity occurs and sound quality deterioration occurs unless the timing of both waveforms is controlled. For example, when a consonant acoustic wave segment having the same pitch and a series formant composite waveform are connected, the pitch cycle instantaneously changes near the connection point unless the timings of the two are accurately controlled. This is because the phases of the two are different.

In addition to this, unless the pronunciation (output) timing of the consonant is accurately controlled, the phonological property is impaired and, for example, "sa" changes to "tsua".

Therefore, in the next embodiment, in order to solve the above-mentioned problem, a label is given to the consonant waveform element, and the waveform timing at the connection point is controlled based on the label.

That is, FIG. 6 is a block diagram of a speech synthesizer according to a third embodiment of the present invention. That is, the voice synthesizer includes a voiced sound generator 14 and a consonant sound waveform generator 17.
The voiced sound generator 14 and the consonant sound waveform generator 17 are connected to a controller 21 for controlling the generation of a voice waveform. The control unit 21 includes a consonant sound waveform generation unit 17
Is connected to a consonant sound waveform label storage unit 18 for storing the label attached to each consonant element stored in the consonant sound waveform storage unit 19 of FIG.
The output of is connected to the output unit 20 in parallel via the combining unit 22. Further, the voiced sound source 15 and the serial formant synthesizer 16 are provided inside the voiced sound generator 15, and the output of the voiced sound source 15 is connected to the serial formant synthesizer 16 and the serial formant synthesizer 16 outputs the same. The output is connected to the synthesizer 22 as the output of the voiced sound generator 14. Here, the voiced sound generation unit 14 described above is a sound generation unit, the consonant waveform storage unit 19 is a waveform segment storage unit, and the consonant waveform label storage unit 18 is a waveform segment feature amount storage unit.

As shown in FIG. 7, the consonant waveform label storage section 18 stores labels representing waveform timings as feature quantities for all necessary consonant segments.
FIG. 7 is an explanatory diagram of a method for assigning a label to unvoiced consonant phonemes. In Fig. 7, strt is the "start label", brst is the "burst label", sov is the "voicing start label", and pea.
k is a "peak label" and end is an "end label". In addition, the values of gain and magn are also stored as the feature amount. gain is "gain information", magn is "peak value information"
Is.

Here, the start label and the end label are literally the start point and the end point of the pronunciation (output) of the consonant element. The end label is attached to the zero-cross point of two pitch periods after the start of vocal cord sound source vibration. This is because the consonant feature included in the subsequent vowel part is included in the consonant segment. To include as many features in the consonant as possible,
We would like to have a large number of pitch periods, but then the pitch of the consonant element itself will be strongly perceived. When the pitch at the time of synthesis is different from this, the sound quality is degraded due to the discontinuity of the pitch. Therefore, in consideration of these, the smallest possible number of pitch periods is selected for each consonant element within a range sufficiently including the features of consonants. A consonant element having a pitch period number of 1 or 2 has a weak pitch perception (referred to as pitch property) and may be used as it is without considering the pitch at the time of synthesis. It is necessary to consider the pitch at the time of synthesis because consonant phonemes and voiced consonants having a larger number of pitch periods have stronger pitch characteristics. Therefore, a consonant element with a plurality of pitches is prepared, and the one with the closest pitch is selected and used during synthesis, or a pitch changing operation is performed on the consonant element (linear expansion / contraction method or pitch synchronization superimposition). Method) method etc. are used.

The burst label is a label attached to the moment when the consonant characterizing each consonant such as the explosive part of the explosive consonant and the frictional part of the fricative consonant (herein, they are collectively referred to as explosive event), and at the time of synthesis. It is used to determine the pronunciation timing of consonant pieces.

The voicing start label is a label attached when a consonant element is an unvoiced consonant. This label is used to synthesize unvoiced consonants. Devoicing is a phenomenon in which trailing vowels of unvoiced consonants disappear due to endings and subsequent phoneme environments. When synthesizing devoiced consonants, consonant segment pronunciation ends with this label. Since the devoicing is essentially the difference between whether or not the vocal cords vibrate after the consonant part, it can be reproduced by stopping the pronunciation at the vocal cord vibration start point in this way.

The peak label is given to the peak on the waveform immediately before the end label and is used for synchronizing the outputs of the voiced sound generator 14 and the consonant waveform generator 17 which will be described later. This peak occurs at the moment the vocal cords close.

When the consonant element is a voiced consonant, a phonological start label is given instead of the voicing start label.
FIG. 8 is an explanatory diagram of a labeling method for a voiced consonant segment. Strt, brst, peak and end are added like unvoiced consonants, but no voicing start label is added. Here sov is used as the phonological start label. The phonological start label is given immediately before the phonological change when the pronunciation start position is gradually delayed from the start label.
This position is generally before the burst label, in the closed section for plosives, and in the section corresponding to the closed section for other phonemes. The closed section is a waveform during which a part of the vocal tract is closed during the production of a plosive sound and the pressure in the vocal tract is increased. Voiced consonants are pronounced at the beginning of a sentence, or at the start label immediately after a pause, otherwise (in the middle of a sentence, such as
In the case where there is no silence or pause immediately before), it is controlled to start from the phonological start label. In this way, it is possible to reproduce the phenomenon that the closed section is shortened in the sentence and to make the consonant unit in the sentence common to the beginning of the sentence.

The gain information is a value for absorbing the difference in the volume of each consonant element and producing a sound at an appropriate volume during synthesis.

The peak value information indicates the amplitude of the peak waveform to which the peak label is added, and is used to smoothly connect the amplitude envelope of the consonant element and the amplitude envelope of the output waveform of the voiced sound generator 14.

The voiced sound source section 15 generates a vocal cord sound source waveform. This waveform is extracted from the actual voice by the inverse filter method. The inverse filter method is a method of extracting the vocal cord source waveform by removing the influence of the vocal tract included in the actual speech waveform, that is, the formant, with a filter having an inverse characteristic of the vocal tract (inverse filter). The waveform thus obtained is called a differential glottal volume flow waveform and corresponds to a waveform obtained by differentiating the acoustic vibration waveform applied to the vocal tract. Thus, this waveform produces a sharp upward pulse at the instant the vocal cords close rapidly. The sharp upward pulse of this waveform is caused by the rapid closure of the vocal cords.

Next, the operation of the speech synthesizer of the above embodiment will be described with reference to the drawings.

First, when the voice to be synthesized is a vowel, the voiced sound source unit 15 generates a vocal cord sound source waveform corresponding to the pitch period. Since the power rises gently at the vowel start portion in natural voice, the voiced sound source section 15 controls the output amplitude to rise at an appropriate time constant. The in-line formant synthesizer 16 adds a formant to this sound source waveform to output it as a vowel.

Next, the case where the voice to be synthesized is a consonant will be described. The output of the consonant sound waveform generation unit 17 and the output of the voiced sound generation unit 14 are used together to synthesize the consonant. First,
Determine the pronunciation timing of consonant pieces. Speech parameters that change from moment to moment are transmitted to the speech synthesizer,
This includes information about phoneme segment switching. For example, the syllable “ka” is divided into “/ k /” segment and “/ a /” segment. The switching of these segments is taken out from the parameter sequence, and the consonant waveform generating section 17 starts the sound generation of the consonant element in advance so that the burst label is matched therewith. By doing so, natural sounding timing of consonants is generated. Further, the consonant waveform generation section 17 controls the output level of the consonant element using the gain information.

After the burst label is sounded, the control unit 2
1 is the voiced sound generation unit 14 until the end label arrives.
Start pronunciation. At this time, the voiced sound source unit 1 is arranged so that the peak label and the peak of the output of the voiced sound source unit 15 match.
The timing of starting the sound generation of No. 5 is controlled. As described above, the upward sharp pulse accompanying the vocal cord closure of the voiced sound source section 15 causes an upward peak on the output waveform of the serial formant synthesis section 16, and as a result, the peak label and the serial formant synthesis section 16 have their respective peaks. The output waveform peaks match.

When one pitch period before the end label comes, the superposition of the outputs of the voiced sound generator 14 and the consonant waveform generator 17 is started. That is, the output of the consonant sound waveform generation unit 17 is attenuated in the section up to the end label with the cosine characteristic, and the output of the voiced sound generation unit 14 is activated with the opposite characteristic. By this operation, the discontinuity on the waveform is removed, but since the consonant sound waveform generation unit 17 and the voiced sound generation unit 14 are synchronized by the peak mark, an extremely smooth waveform connection without fluctuation of the pitch period can be achieved. Will be realized.

At the same time, by controlling the time constant for raising the output amplitude of the voiced sound source section 15, the amplitude envelopes of the outputs of the voiced sound generation section 14 and the consonant waveform generation section 17 are smoothly connected. The peak value information is used for this control. That is,
The time constant may be determined so that the amplitude of the voiced sound generation unit 14 at the time of the peak label becomes the value represented by the peak value information. Since the peak value can be obtained by reading the value of the consonant element at the time of the peak label, it does not have to be stored in the consonant waveform label storage unit 18.

Next, FIG. 9 shows how the waveform connection is made. Figure 9
Is the output waveform of the voiced sound source unit 15 from above, and the voiced sound generation unit 14
The output waveform of the (series formant synthesis unit 16), the output waveform of the consonant waveform generation unit 17, and the output waveform (synthesis waveform) of the output unit 20 are shown. In FIG. 9, the broken line drawn over all four waveforms represents the peak label of the consonant element. By synchronizing the peak of the voiced sound source unit 15 with the peak label of the consonant element, it can be seen that the output of the voiced sound generation unit 14 is connected to the consonant element at an appropriate timing.

The same waveform connection method can also be used when connecting a consonant element after the output waveform of the voiced sound source section 15. When the consonant element is a voiced consonant, a peak label is given to a peak on the waveform immediately after the start of the consonant element, and the peak label is controlled so as to be synchronized with the peak of the output waveform of the preceding voiced sound source section 15. You can make a smooth connection.

As described above, in order to prevent the waveform discontinuity and the pitch fluctuation at the connection point, the consonant element is labeled in advance and the output of the voiced sound generator 14 and the consonant waveform generator 17 is used as a clue. It is for synchronization. Also, in order to eliminate the need to prepare a dedicated consonant element for devoicing, a label is given to a normal consonant element that is not devoiced, and the devoicing is reproduced by using the label during synthesis. Is. Then, by using the phonological start label, it is possible to synthesize by using consonant phonemes common to the beginning of a sentence and a sentence.

As a result, the output of the voiced sound generation unit 14 and the output of the consonant sound waveform generation unit 17 are connected smoothly and at appropriate timings, and high-quality speech without noise or discontinuity in pitch can be synthesized. . Further, it is not necessary to prepare a dedicated waveform segment for devoicing, the beginning of a sentence, and the middle of a sentence, and a common consonant segment can be used, so that the storage capacity and the recording time can be reduced.

In the above embodiment, the case where the consonant segment is used as the waveform segment has been described, but the waveform segment to be used may of course be any other phoneme.

In the above embodiment, the control unit 21 controls the sounding start timing of the voiced sound source unit 15 in order to match the peak positions on the waveform. However, the present invention is not limited to this. Output waveform and child sound waveform generator 17
It is also possible to control either one or both of the pronunciation times of.

Further, in the above-mentioned embodiment, each processing unit is constituted by dedicated hardware, but instead of this, the same function may be realized by software using a computer.

So far, the construction method for synthesis of unvoiced consonants, voiced fricatives, voiced plosives, etc. has been described. For the phonemes with feature parameters such as nasal sounds for a considerably long time, the above consonant is used. Sufficient sound quality cannot be obtained with one configuration. As described above, the length of the segment must be sufficiently short in order to make the connection without considering the pitch. However, it is impossible to include long-term changes in characteristic parameters such as nasal sounds in such short segments. In addition to nasal sounds, there are many phonemes that have long characteristic parameters up to the following vowel part, and for these, the sound quality can be expected to be improved by lengthening the segment length without considering articulation coupling.

When the segment length is increased, the segment and the series formant synthesized waveform are connected near the center of the vowel.
Since the vicinity of the center of the vowel is a relatively stationary part where the spectrum change is small, the rapid spectrum change due to the connection has a great influence on the sound quality. In order to solve this problem, it is effective to perform the superposition processing at the connection point in a longer section.

However, when the pitch of the segment is different from the synthetic pitch in the overlapping section, both waveforms interfere with each other, and an echo or noise is generated. In addition, since the long piece itself has a strong pitch property, pitch discontinuity before and after the connection is large and the sound quality is impaired.

Therefore, it is conceivable to prepare consonant phonemes having various pitches according to the synthetic pitch.
In order to perform pitch alignment with sufficiently high precision, it is necessary to prepare an extremely large number of kinds of pieces. Further, the synthetic pitch changes depending on the intonation pattern, and the change greatly occurs within the duration of the consonant element. As described above, it is virtually impossible to prepare consonant phonemes corresponding to various pitch changes.

Therefore, it becomes indispensable to add a pitch changing operation to the prepared consonant element. As a simple pitch changing method, there is a linear expansion / contraction method. This method is a method for obtaining a waveform expanded or contracted along the time axis by reading out the stored waveforms one by one in order when reading the stored waveforms at intervals other than 1. Since the read address of the stored waveform becomes a non-integer address that does not actually exist due to the non-integer interval, interpolation is performed using straight lines from the preceding and following values.

However, even if the pitches match in the overlapping section, it is difficult to accurately synchronize the phases. This is because the linear expansion / contraction method is a method of changing the pitch at a constant rate based on the original pitch, and therefore must have extremely accurate information on the original pitch and its fluctuation. Therefore, it can be said that the waveform synchronization method according to the above-described embodiment cannot achieve phase synchronization for a long time. In addition, since the pitch changing operation by linear expansion / contraction involves a change in spectrum shape, it causes problems such as sound quality deterioration, phonological deterioration, and spectrum discontinuity due to connection. For this reason, the pitch can be changed only within a range extremely smaller than the original pitch.

Therefore, in the next embodiment, in order to solve the above-mentioned problem, the pitch synchronization superposition method is used, and the method of always obtaining the phase synchronization of the waveform by using the pitch synchronization signal is adopted.

FIG. 10 is a block diagram of a speech synthesizer according to the fourth embodiment of the present invention. The voice synthesizer is provided with a pitch control unit 1, and outputs the pitch control signal generation unit 24 and the waveform reading units 26a, 26b, 26c, 26.
d, and is connected to the window hanging portions 28a, 28b, 28c, 28d. The output of the pitch synchronization signal generation unit 24 is connected to the pitch synchronization signal distribution unit 24a and the delay unit 37. The first output of the pitch synchronization signal distribution unit 25a is to the waveform reading unit 26a, and the second output is the waveform reading unit 26b.
Have been entered respectively. The output of the delay unit 37 is input to the pitch synchronization signal distribution unit 25b, the first output thereof is the waveform reading unit 26c, and the second output thereof is the waveform reading unit 2.
6d is input respectively.

The outputs of the voiced sound source waveform storage unit 27 and the offset control unit 41 are connected to the waveform reading units 26a and 26b. The output of the voiced sound source peak position storage unit 29 is connected to the input of the offset control unit 41. The output of the waveform reading unit 26a is input to the windowing unit 28a, and the output of the waveform reading unit 26b is input to the windowing unit 28b. The output of the window unit 28a is connected to the mixing unit 31a. The output of the windowing section 28b is connected to the mixing section 31a via the gain control section 30. The output of the mixing unit 31a is input to the series-type formant synthesis unit 32 via the gain control unit 40a.

The outputs of the consonant sound waveform storage unit 33, the consonant sound waveform peak position storage unit 34, and the consonant sound waveform label storage unit 42 are connected to the waveform reading units 26c and 26d, and the output of the waveform reading unit 26c is sent to the windowing unit 28c. The output of the waveform reading unit 26d is input to the windowing unit 28d. Both outputs of the window hanging portion 28c and the window hanging portion 28d are input to the mixing portion 31b. The output of the mixing section 31b is connected to the gain control section 40b.

The outputs of the series formant combiner 32 and the gain controller 40b are connected to the combiner 35, and the outputs thereof are connected to the output 36.

Next, the operation of the speech synthesizer configured as above will be described.

The F0 parameter generated by the pitch control unit 23 in accordance with the intonation pattern is the pitch synchronization signal generation unit 24 and the waveform reading units 26a, 26b, 26c, 2
6d, transmitted to the window hanging portions 28a, 28b, 28c, 28d. The pitch synchronization signal generation unit 24 generates a pitch synchronization signal having a cycle according to the F0 parameter and outputs it to the pitch synchronization signal distribution unit 25a and the delay unit 37.

First, a method of generating a voiced sound source using the pitch synchronization superposition method will be described.

The pitch synchronizing signal distributor 25a converts the input pitch synchronizing signal into two waveform reading units 26a and 26b.
Alternately output to.

When receiving the pitch synchronization signal, the waveform reading section 26a reads the first peak position from the voiced sound source peak position storage section 29 through the offset control section 41. The offset control unit 41 adds the offset Noff to the output of the voiced sound source peak position storage unit 29 and outputs the result. Noff
Will be described later. The waveform reading unit 26a starts reading the voiced sound source waveform stored in the voiced sound source waveform storage unit 27 based on the thus obtained peak position with offset. The read start position N0 is given by (Equation 1).

[0072]

## EQU1 ## N0 = P0-Noff-Tsyn Here, P0 is the 0th peak position stored in the voiced sound source peak position storage unit 29, and Tsyn is a synthetic pitch period based on the F0 parameter.

The output of the waveform reading section 26a is the windowing section 2
8a, and windowing is performed by the Hanning window. The length of the Hanning window Twin is twice the smaller of the synthetic pitch period Tsyn and the original pitch period Torg of the voiced sound source waveform. This is to prevent deterioration of sound quality due to the peaks on both sides entering the Hanning window when Twin exceeds twice Torg. In this way, the pitch waveform is generated.

The pitch synchronizing signal is transmitted to the waveform reading section 26b one pitch cycle later than this operation. The waveform reading unit 26b reads the waveform as before, and the windowing unit 28b performs windowing. The waveform reading start position at this time is given by (Equation 2).

[0075]

N1 = P1−Noff−Tsyn where P1 is the first peak position stored in the voiced sound source peak position storage unit 29.

The output of the windowing section 28b is subjected to gain control in the gain control section 30 in the range of 0 to 1. The purpose is to simulate the unstable vocal cord vibration that occurs at the beginning and end of words. That is, at the beginning and end of a word, the vocal cords may repeat large and small vibrations every pitch period, resulting in a double pitch period component. By setting the gain to 0.5 or the like in the gain controller 30, it is possible to generate a double pitch period component.

By superposing the pitch waveforms alternately generated as described above in the mixing section 31a,
A voiced sound source waveform having a desired pitch period is generated.
Further, since the individual pitch waveforms are not expanded / contracted with respect to the time axis, the spectral shape does not change.

The voiced sound source waveform thus generated is subjected to amplitude control by the gain control section 40a and then articulated by the conventional series formant synthesis section 32 to become a vowel component.

Next, the above Noff will be described. When the pitch of the voiced sound source waveform is changed, spectrum distortion may occur due to the following reasons. The glottal volume flow waveform extracted by the inverse filter method has a structure as shown in FIG. Of these, the open-glottic waveform has low-range energy, and the closed-glottic waveform has high-range energy.

FIG. 12 is a diagram when the pitch frequency is changed to be lower than the original pitch frequency under Noff = 0. Since the glottal closure is located near the edge of the Hanning window, it decays when the overlapping section of adjacent Hanning windows becomes shorter. For this reason, the voiced sound source waveform generated has low-frequency energy components.

In order to prevent this, as shown in FIG. 13, the glottic closure is shifted by Noff samples from the center of the Hanning window so that the glottal opening approaches the center of the Hanning window. However, if Noff is made too large, the pulse-like waveform at the glottal blockage is attenuated when the pitch is raised, and the high-frequency energy is reduced. This is because when the pitch frequency is changed to be higher than the original pitch frequency, the Hanning window length becomes shorter, so that the glottal closing pulse approaching the edge of the Hanning window is attenuated. For this reason Noff is 0.1To
Use degree.

In the consonant generation process, waveform reading and windowing are performed as in the voiced sound source. The pitch synchronizing signal which is the input thereof is delayed by Noff samples by the delay unit 37. As a result, the peak position of the consonant sound waveform and the peak position of the voiced sound source waveform are synchronized. Further, similarly to the third embodiment, the sound generation timing is controlled according to the consonant sound waveform label storage unit 42.

The vowel component waveform and the consonant component waveform thus generated in synchronization with each other are smoothly superposed on each other in the synthesizing section 35, converted into voice by the output section 36 and output. As a result, it is possible to obtain an extremely high-quality synthesized sound that does not have waveform discontinuity, pitch discontinuity, or phase discontinuity using the waveform segment in the consonant part.

In the present embodiment, a single voiced sound source waveform is used for the voiced sound source section, but it is possible to obtain a higher quality synthesized speech using a plurality of sound source waveforms by simple extension. For example, it is conceivable to mix a sound source with many harmonic components and a sound source with few harmonic components in some cases, or prepare a dedicated sound source for the five vowels and combine them while switching.

FIG. 14 is a block diagram of a speech synthesizer according to the fifth embodiment of the present invention. The voice synthesizer has a configuration in which five voiced sound source units 38 in the fourth embodiment are provided. That is, the pitch control unit 1 is provided, and the outputs thereof are the pitch synchronization signal generation unit 24, the voiced sound source units 38a, 3 and 3.
8b, 38c, 38d, and 38e. The output of the pitch synchronization signal generator 24 is the pitch synchronization signal distributor 2
5a and the delay unit 37. The two outputs of the pitch synchronization signal distribution unit 25a are voiced sound source units 38, respectively.
Each of a, 38b, 38c, 38d, and 38e is connected to two inputs. Voiced sound source section 38
Voiced sound sources are generated inside a, 38b, 38c, 38d, and 38e as in the fourth embodiment, and the outputs thereof are mixed and input to the serial formant synthesis unit 32.

On the other hand, the output of the delay unit 37 is connected to the pitch synchronization signal distribution unit 25b. The two outputs of the pitch synchronization signal distributor 25b are connected to the consonant generator 39. Inside the consonant generation unit 39, consonant components are generated using consonant sound waveform segments as in the fourth embodiment.

The outputs of the serial formant synthesis unit 32 and the consonant generation unit 39 are input to the synthesis unit 35, and the output of the synthesis unit 35 is input to the output unit 36.

In the five voiced sound source sections 38a to 38e, glottal volume flow waveforms extracted from the five vowels / a / to / o / by the inverse filter method are stored. The sound source waveform extracted by the inverse filter method is slightly different depending on the five vowels. Therefore, it is possible to synthesize high-quality speech by performing synthesis of the five vowels from the sound source waveform extracted from each of the five vowels, rather than performing synthesis from the common sound source waveform.

Therefore, the sound quality of each vowel can be improved by switching these sound sources at vowel and syllable breaks. Gain control unit 40a when switching
Noise and abnormal noise can be suppressed by smoothly moving the gains of the sound sources up and down. Since each sound source is accurately peak-synchronized, an extremely natural sound source waveform can be generated even if such superposition and switching are performed.

Since the original pitches of the sound sources of the five vowels are different from each other and each pitch includes fluctuations, it is extremely difficult to achieve complete synchronization by the structure of the voiced sound source section using the conventional linear expansion / contraction. However, according to the configuration of the present invention, the original pitch of each sound source may be different, and the pitch may include fluctuation.

In this embodiment, the voiced sound source section is made plural for the five vowels, but it may be made plural according to another standard. For example, there are pluralization by a sound source having many harmonics and plural sound sources with few harmonics, pluralization by a pitch range, pluralization by a position in a sentence (beginning of sentence, in the sentence, end of sentence, etc.).

In the present embodiment, the pitch synchronization signal common to all voiced sound source sections and consonant sections is used for synchronization.
It is also possible to calculate the pitch period in each part based on the F0 parameter and read the waveform. In this case, it is sufficient to synchronize with each other at the start of sound generation.

Although the window function is a Hanning window having a length twice as long as the smaller one of the synthetic pitch period and the original pitch period, a window having another shape or length may be used.

[0094]

As is apparent from the above description,
According to the present invention, a vowel signal is flexible and can give various tones and intonation by a serial formant synthesis method, and a consonant signal is a high-quality consonant that cannot be realized by a formant synthesis method using a waveform segment. Therefore, the synthesized sound obtained by combining them can have high quality and can correspond to various voice qualities. Further, in contrast to the conventional method using the waveform element, in the case of the present method, since the storage as the waveform element is limited to the consonant having a short duration, it can be realized by a small-capacity storage device.

Further, by providing the unvoiced sound source section and the parallel type formant synthesizing section, the parallel type formant synthesizing section can be used for the consonant which can achieve sufficiently high quality by the formant synthesizing method, and can store the waveform segment. It is possible to further reduce the storage capacity required for. In addition, by using the parallel formant synthesizer and the waveform element at the same time, the characteristics of the waveform element itself can be changed, which is effective in securing clarity in a telephone band or in a noise environment.

Further, the present invention comprises control means for synthesizing the voice signal generated by the sound generating means and the waveform element signal obtained from the waveform element storage means based on the characteristic amount of the waveform element. Therefore, it has advantages that it is possible to suppress the generation of noise due to the connection of the voice waveform, and it is possible to reduce the storage capacity for storing the waveform element and the recording work.

Further, according to the present invention, the pitch synchronization superposition method is used for the pitch control of the voiced sound source section and the consonant generation section, whereby the voiced sound source waveform and the consonant sound waveform are perfectly synchronized, and the waveform discontinuity, pitch discontinuity, It is possible to synthesize extremely high quality speech without phase discontinuity. Further, it is possible to avoid a change in the spectrum shape due to the pitch change.
Furthermore, voiced sound sources having a plurality of different characteristics can be mixed or switched according to the purpose, and high-quality speech can be synthesized using appropriate sound sources according to various situations.

[Brief description of drawings]

FIG. 1 is a block diagram of a speech synthesizer according to a first embodiment of the present invention.

FIG. 2 shows a state in which a waveform segment of a consonant “/ k /” and a synthesized signal of a vowel “a” are combined into a “ka”.

FIG. 3 is a block diagram of a voice synthesizing apparatus having a parallel formant synthesizer in a second embodiment of the present invention.

FIG. 4 is a block diagram of a conventional formant type speech synthesizer.

FIG. 5 is a block diagram of a conventional speech synthesizer using a waveform segment.

FIG. 6 is a block diagram of a speech synthesizer according to a third embodiment of the present invention.

FIG. 7 is a diagram illustrating labeling of unvoiced consonant segments in the example.

FIG. 8 is a diagram illustrating labeling of voiced consonant segments in the example.

FIG. 9 is a diagram illustrating waveform connection in the same example.

FIG. 10 is a block diagram of a speech synthesizer according to a fourth embodiment of the present invention.

FIG. 11 is a diagram illustrating a glottal volume flow waveform.

FIG. 12 is a diagram illustrating an operation of lowering the pitch frequency by the pitch synchronization superposition method.

FIG. 13 is a diagram illustrating the positional relationship between the Hanning window and the glottal volume flow waveform of the present invention.

FIG. 14 is a block diagram of a speech synthesizer according to a fifth embodiment of the present invention.

[Explanation of symbols]

 1 voiced sound source section 2 serial formant synthesis section 3 consonant waveform storage section 4 consonant waveform readout section 5 synthesis section 6 unvoiced sound source section 7 parallel formant synthesis section 8 formant synthesis control coefficient generation section 9 formant synthesis control rule storage section 10 formant synthesizer 11 speech unit selection unit 12 speech unit database storage unit 13 unit connection synthesis unit 14 voiced sound generation unit 15 voiced sound source unit 16 series formant synthesis unit 17 consonant sound waveform generation unit 18 consonant sound waveform label storage unit 19 Consonant waveform storage unit 20 Output unit 21 Control unit 22 Synthesizing unit 23 Pitch control unit 24 Pitch synchronization signal generation unit 25 Pitch synchronization signal distribution unit 26 Waveform reading unit 27 Voiced sound source waveform storage unit 28 Windowing unit 29 Voiced sound source peak position storage unit 30 Gain Control Section 31 Mixing Section 32 Serial Formant Synthesis Section 33 Consonant Sound Waveform Part 34 consonant waveform peak position memory 35 combining unit 36 output unit 37 delay unit 38 a voiced sound source unit 39 consonant generator 40 gain control section 41 offset control unit 42 consonant waveform label storage unit

Claims (24)

[Claims] 1. A voiced sound source section for outputting a voiced sound source signal, and a plurality of formant resonators connected in series with the voiced sound source signal from the voiced sound source section as an input, for synthesizing a predetermined sound such as a vowel. A serial formant synthesis unit, a waveform storage unit that stores a waveform of a predetermined sound such as a plurality of consonants, a waveform reading unit that reads out a required waveform from the waveform storage unit, and a serial formant synthesis unit. A speech synthesizer, comprising: a waveform combining section that outputs the synthesized speech by superimposing or switching the waveform read by the waveform reading section and outputting the synthesized speech. 2. A voiced sound source section for outputting a voiced sound source signal, and a plurality of formant resonators connected in series with the voiced sound source signal from the voiced sound source section as an input to synthesize a predetermined sound such as a vowel. A serial formant synthesis unit, a waveform storage unit that stores a waveform of a predetermined sound such as a plurality of consonants, a waveform readout unit that reads out a required waveform from the waveform storage unit, and an unvoiced sound such as white noise An unvoiced sound source unit, a sound source signal from the unvoiced sound source unit as an input, and having a plurality of resonators connected in parallel, a parallel formant synthesis unit that synthesizes a predetermined sound such as a plosive sound or a fricative sound, and A waveform combination unit that outputs the synthesized voice by superimposing or switching the output of the serial formant synthesis unit, the output of the parallel synthesis unit, and the waveform read by the consonant sound waveform reading unit. A speech synthesizer characterized by the above. 3. A sound generating means for generating a voice signal based on a characteristic parameter extracted from a voice, a waveform element storage means for storing a waveform element cut out from the voice, and the stored waveform element. A waveform segment feature amount storage means for storing a predetermined feature amount for each piece, and a voice signal generated by the sound generation means and the waveform segment storage means based on the stored feature amount of the waveform segment. A voice synthesizing apparatus comprising: a control means for synthesizing the obtained waveform segment signal. 4. The speech synthesizer according to claim 3, wherein the predetermined feature amount is a gain value of the waveform segment, and the amplitude of the waveform segment signal is controlled by the gain value. . 5. The voice according to claim 3, wherein the waveform segment feature amount storage means stores the existence time of a waveform having a desired feature on the waveform segment. Synthesizer. 6. The voice synthesizing apparatus according to claim 5, wherein the desired feature is any peak position or peak value on the waveform of the waveform segment. 7. The sound generation means has a voiced sound source generation section for generating a voiced sound source, and the control means sets the peak position of the output waveform of the sound generation means at the peak position on the waveform of the waveform segment. 7. One or both of the phase of the output waveform of the voiced sound source generation unit and the sounding (output) timing of the waveform segment are controlled so that they coincide with each other.
The described speech synthesizer. 8. The driving of the voiced sound source generation unit is started so that the peak of the output waveform of the voiced sound source generation unit coincides with the peak position on the waveform of the waveform segment. 7. The speech synthesizer according to 7. 9. The amplitude of the voiced sound source generation section is controlled so that the amplitude envelope of the output of the sound generation means becomes a peak amplitude value at the peak position. Speech synthesizer. 10. The waveform segment is cut out from a start portion of a consonant up to a predetermined number of pitch periods of a subsequent vowel, and is formed, as claimed in claim 3, claim 4, or claim 5.
Alternatively, the speech synthesizer according to claim 6 or claim 7 or claim 8 or claim 9. 11. The desired feature is the consonant timing of each consonant such as a plosive event when the consonant element is a plosive sound and a friction event when the consonant element is a plosive sound.
The described speech synthesizer. 12. The speech synthesizer according to claim 11, wherein the sound generation of the consonant element is started in advance with reference to the timing of articulation. 13. The speech synthesizer according to claim 5, wherein when the consonant element is an unvoiced consonant, the desired characteristic is the time of existence of the vocal cord vibration start event of the unvoiced consonant. 14. The speech synthesizer according to claim 11, wherein when synthesizing a devoiced consonant, the pronunciation of a consonant element is stopped by using the position of the vocal cord vibration start event. 15. When the consonant segment is a voiced consonant, the desired feature is the position of the phonological onset event, which is the position at which the phonological feature does not change even if the waveform before that position is removed. The speech synthesizer according to claim 5, characterized in that: 16. The speech synthesis apparatus according to claim 15, wherein when the consonant to be pronounced is not silent or pause immediately before, the pronunciation is started from the time of existence of the phonological start event. 17. A voiced sound source waveform generating means for generating a voiced sound, a serial formant synthesizer, a consonant sound waveform generating means for generating a consonant, a waveform connecting means for connecting a waveform, and a pitch synchronizing signal generating means. The synchronizing signal generating means outputs a pitch synchronizing signal corresponding to a desired pitch period, and the voiced sound source waveform generating means and the consonant sound waveform generating means both generate a waveform having a phase synchronized with the pitch synchronizing signal, and the serial type The formant synthesis unit changes the frequency characteristic by a transfer function simulating the vocal tract characteristic to the output waveform of the voiced sound source waveform generation unit, and the waveform connection unit generates the output waveform of the serial formant synthesis unit and the consonant waveform generation unit. A speech synthesizer characterized by generating a speech waveform by connecting or mixing the output waveforms of the means. 18. A voiced sound source peak, comprising pitch synchronization signal distribution means, wherein said voiced sound source waveform generation means stores a voiced sound source waveform storage means and a peak position on the voiced sound source waveform stored in said voiced sound source waveform storage means. The position synchronization means, the first pitch waveform cutting means, the second pitch waveform cutting means, and the mixing section are provided, and the pitch synchronization signal distribution means distributes the pitch synchronization signals alternately into two parts, respectively. The first
Output to the second pitch waveform cutting means and the first pitch waveform cutting means and the second pitch waveform cutting means from the voiced sound source waveform storage means to the voiced sound source peak position storage means. A pitch waveform cut out by a window function in which the window length is about twice the desired pitch period and the both ends are focused to near zero around the peak position stored in is immediately after the distribution pitch synchronization signal is received. 18. The speech synthesizer according to claim 17, wherein the speech is output to a mixing section, and the mixing section mixes outputs of the first pitch waveform cutting means and the second pitch waveform cutting means. 19. The voice synthesizing apparatus according to claim 18, wherein the voiced sound source generation means includes a gain control means, and controls the gain of either one of the two waveforms input to the mixing means. 20. Pitch synchronization signal distribution means, said consonant sound waveform generation means, a plurality of consonant sound wave shape storage means, a plurality of consonant sound wave peak position storage means corresponding to said plurality of consonant sound wave shape storage means, and a first pitch. Waveform cutting means and second
Pitch waveform cutting means and mixing means, the pitch synchronizing signal distributing means alternately distributes the pitch synchronizing signal into two, and each of the pitch synchronizing signals is divided into two parts.
Output to the second pitch waveform cutting means and the second pitch waveform cutting means, wherein the first pitch waveform cutting means and the second pitch waveform cutting means are consonant waveforms corresponding to desired consonants from the consonant waveform storage means. Is a window function that focuses on the peak position corresponding to the desired consonant stored in the consonant waveform peak position storage means and has a window length of about twice the desired pitch period and both shorts converged near zero. The cut-out pitch waveform is output to the mixing unit immediately after receiving the distributed pitch synchronization signal, and the mixing unit mixes the outputs of the first windowing unit and the second windowing unit. 18. The speech synthesizer according to claim 17, which is characterized in that. 21. Pitch synchronization signal distributing means, wherein the consonant sound waveform generating means comprises a plurality of consonant sound waveform storage means, a plurality of consonant sound waveform peak position storage means corresponding to the plurality of consonant sound waveform storage means, and a first pitch. Waveform cutting means and second
Pitch waveform cutting means and mixing means, the pitch synchronizing signal distributing means alternately distributes the pitch synchronizing signal into two, and each of the pitch synchronizing signals is divided into two parts.
Output to the second pitch waveform cutting means and the second pitch waveform cutting means, wherein the first pitch waveform cutting means and the second pitch waveform cutting means are consonant waveforms corresponding to desired consonants from the consonant waveform storage means. Is a window function that focuses on the peak position corresponding to the desired consonant stored in the consonant waveform peak position storage means and has a window length of about twice the desired pitch period and both shorts converged near zero. The cut-out pitch waveform is output to the mixing unit immediately after receiving the distributed pitch synchronization signal, and the mixing unit mixes outputs of the first windowing unit and the second windowing unit. The speech synthesizer according to claim 18, which is characterized in that. 22. Pitch synchronization signal delay means, said voiced sound source waveform generation means comprises offset control means, said offset control means advancing the reading start position of pitch waveform cutting means by the offset value The pitch of the voiced sound source waveform with respect to the center of the voiced sound source waveform, the pitch synchronization signal delaying means delays the pitch synchronization signal by the offset value, and delays the output of the consonant waveform generation means by the offset to delay the voiced sound source waveform. 22. The speech synthesizer according to claim 21, wherein 23. A method according to claim 18, further comprising a plurality of voiced sound source generation means, wherein all the voiced sound source generation means perform synchronization by using a common pitch synchronization signal or distributed pitch synchronization signal. Alternatively, the speech synthesizer according to claim 21. 24. A voice synthesis apparatus according to claim 22, further comprising a plurality of voiced sound source generation means, wherein all the voiced sound source generation means perform synchronization using a common pitch synchronization signal or distributed pitch signal and an offset value.