JPH09160595A - Voice synthesizing method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a voice synthesizing method suited for obtaining a high- quality synthesized voice in text-to-speech synthesis. SOLUTION: This method produces synthesized voice signals by connecting together the pieces of information that are selected from plural pieces of prestored voice synthesis unit information. In this case, a formant emphasizing filter part 17, in which an LPC that is a voice spectrum parameter for use as a coefficient in a vocal-tract filter part 16, is used as a filter factor 112, is provided, and the formant of the synthesized voice signals is emphasized by the filter 17.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech synthesis method for generating a synthesized speech signal from information such as a phoneme symbol string, pitch and phoneme duration in a text-to-speech system.

[0002]

2. Description of the Related Art Artificially producing a voice signal from an arbitrary sentence (text) is called text-to-speech synthesis. Usually, a text-to-speech synthesis system is composed of three elements: a language processing section, a phoneme processing section, and a speech signal generation section. The input text is first subjected to morphological analysis and syntactic analysis in the language processing unit, and then subjected to accent and intonation processing in the prosody processing unit,
Information such as a phoneme symbol string, pitch, and phoneme duration is output. Finally, the voice signal generator, that is, the voice synthesizer, synthesizes a voice signal from information such as a phoneme symbol string, pitch, and phoneme duration. Therefore, the speech synthesis method used for text-to-speech synthesis must be a method capable of synthesizing an arbitrary phoneme symbol string.

The principle of a speech synthesis system capable of synthesizing such a phoneme symbol sequence is as follows.
The characteristic parameter, which is information of a basic voice synthesis unit such as a pitch section, is connected by controlling the pitch and duration. A vocoder system and a formant synthesis system are conventionally known as a system of a speech synthesis apparatus capable of synthesizing an arbitrary phoneme symbol string by controlling a pitch and a duration. These methods synthesize voices by driving a vocal tract filter that models vocal tract characteristics with a drive signal that models vocal cord signals, but the modeling accuracy is insufficient. Therefore, the synthesized voice was unclear.

Therefore, as a method for improving the sound quality by increasing the modeling accuracy, for example, Japanese Patent Application Laid-Open No. 58-80699.
As disclosed in No. "Voice Synthesis Method", it is obtained by controlling a vocal tract filter based on a spectral parameter obtained by analyzing natural speech and processing the speech signal with an inverse filter of the vocal tract filter. There is a method in which the residual waveform is used as a drive signal for the vocal tract filter.

FIG. 17 shows the configuration of a conventional speech synthesizer using the LPC method of residual driving which is an example of this method. This speech synthesizer includes a residual waveform storage unit 11, a voiced sound source generation unit 12, an unvoiced sound source generation unit 13, an LPC coefficient storage unit 14, an LPC coefficient interpolation unit 15, and a vocal tract filter unit 1.
6 is comprised.

The residual waveform storage unit 11 stores a plurality of residual waveforms in advance as information on a plurality of voice synthesis units, and one pitch cycle selected from the residual waveforms according to the waveform selection information 101. The long residual waveform 102 is output. The voiced sound source generation unit 12 repeats the one-pitch cycle length residual waveform 102 with the frame average pitch 103 as a cycle, and multiplies the repeated waveform by the frame average power 102 to generate the voiced sound source signal 105. The voiced sound source signal 105 is output in the voiced section discriminated by the voiced / unvoiced discrimination information 107 and input to the vocal tract filter unit 16. The unvoiced sound source generation unit 13 outputs an unvoiced sound source signal 106 represented by white noise or the like based on the frame average power 102. The unvoiced sound source signal 106 is output in the unvoiced section determined by the voiced / unvoiced determination information 107 and input to the vocal tract filter unit 16.

The LPC coefficient storage unit 14 stores a plurality of LPC coefficients which are information of another speech synthesis unit,
One LPC coefficient 109 according to the C coefficient selection information 108
Is selectively output. The LPC coefficient interpolating unit 15 interpolates the LPC coefficient of the previous frame and the LPC coefficient 109 of the current frame so that the LPC coefficient does not become discontinuous between frames and outputs the LPC coefficient 110.

The vocal tract filter section 16 includes a voiced sound source signal 10
5 or the unvoiced sound source signal 106, the LPC coefficient 110
Driving a vocal tract filter with a coefficient of
Outputs 1.

In this speech synthesizer, various LPC coefficients obtained by performing linear prediction analysis on natural speech in advance are used as LP.
The C-coefficient storage unit 14 stores the waveform of one pitch period from the residual waveform obtained by performing inverse filtering with these LPC coefficients, and stores it in the residual waveform storage unit 11. In this way, since parameters such as LPC coefficients obtained by analyzing natural speech are applied to the vocal tract filter and the sound source signal, the accuracy of modeling is high and a synthetic speech relatively close to natural speech can be obtained. You can

[0010]

However, in the above-mentioned conventional speech synthesizer, even if the modeling is performed with high precision, the natural speech analyzed when obtaining the LPC coefficient and the residual waveform is the pitch period. It is inevitable that spectrum distortion will occur when different voices are synthesized.

For example, if the spectrum envelope of the voice of a certain phoneme is expressed as shown in FIG. 13A, the power spectrum of the voice signal when the phoneme is generated at the fundamental frequency f is shown in FIG. 13B. As shown in (1), a discrete spectrum is obtained by sampling the spectrum envelope at frequency intervals f. Similarly, the power spectrum of the voice signal when uttered at the fundamental frequency f'is a discrete spectrum in which the spectrum envelope is sampled at the frequency interval f ', as shown in FIG. 13 (c).

Here, FIG. 13 uttered at the fundamental frequency f.
By analyzing the voice having the spectrum shown in (b) to obtain the spectrum envelope, the LPC coefficient storage unit 1
Consider finding the LPC coefficient stored in 4. In the case of a voice signal, it is generally impossible in principle to obtain a true spectrum envelope as shown in FIG. 13A from a discrete spectrum as shown in FIG. Therefore, even if the spectrum envelope obtained by analyzing the voice becomes equal to the true spectrum envelope at the discrete points as shown by the broken line in FIG. 14A, an error may occur at other frequencies. There is. Therefore, the obtained spectrum envelope may be a spectrum in which the peak portion (formant) is distorted with respect to the true spectrum envelope, as shown in FIG. In this case, the spectrum of the synthesized speech obtained by performing speech synthesis at a fundamental frequency f ′ different from f is as shown in FIG.
The spectrum becomes blunt in comparison with the spectrum of natural speech shown in FIG. 3 (c), which causes deterioration of clarity of synthesized speech.

Further, when the voice synthesis units are connected, interpolation of parameters such as a filter coefficient causes the unevenness of the spectrum to be averaged and blunted, and the synthesized voice becomes unclear. . For example,
If the frequency characteristics of the LPC coefficients of two consecutive speech synthesis units are represented as shown in FIGS. 15A and 15B, respectively, the frequency of the filter obtained by interpolating these two filter coefficients is shown. The characteristics are shown in Figure 15.
As shown in (c), the unevenness of the spectrum may be averaged and dulled, which may also cause deterioration of the clarity of the synthesized speech.

Further, when the peak position of the residual waveform differs for each frame, there is a problem that the pitch of the voiced sound source is disturbed. For example, even if the residual waveforms are arranged at equal intervals T as shown in FIG. 16, if the positions of the peaks of the residual waveforms are different, the harmonics of the pitch of the synthesized speech signal are disturbed, which causes deterioration in sound quality. .

The present invention has been made to solve the above problems, and an object of the present invention is to provide a speech synthesis method suitable for obtaining high-quality synthesized speech in text speech synthesis.

[0016]

In order to achieve the above-mentioned object, the present invention is a speech synthesis method for generating a synthesized speech signal by connecting information selected from a plurality of pieces of speech synthesis unit information stored in advance. The main point is to provide a formant enhancement filter whose filter coefficient is determined according to the spectral parameter of the voice used as the filter coefficient of the vocal tract filter, and to enhance the formant of the synthesized voice signal by this filter.

That is, the first voice synthesizing method according to the present invention is a voice synthesizing method for generating a synthetic voice signal by connecting information selected from a plurality of prestored information of voice synthesizing units, and storing the voice synthesizing method in advance. The information of the synthesized speech synthesis unit includes at least the spectrum parameter of the speech, and the filter coefficient is determined according to the selected spectrum parameter. By enhancing the formant of the synthesized speech signal by the formant enhancement filter, the rounded spectrum is shaped and becomes clear. It is designed so that various synthetic voices can be obtained.

In a second voice synthesis method according to the present invention, the information of the voice synthesis unit stored in advance includes at least a voice spectrum parameter and a vocal tract filter drive signal of one pitch period, and the filter coefficient is selected according to the selected spectrum parameter. By enhancing the spectrum of the formants of the synthesized speech signal by the formant enhancement filter, the clear synthesized speech can be obtained with a smaller amount of calculation.

A third voice synthesis method according to the present invention includes a formant enhancement process at the time of voice synthesis by including a waveform in which a formant of a waveform of at least one pitch period of a voice is emphasized in information of a voice synthesis unit stored in advance. It is possible to obtain a clear synthetic speech without performing.

In a fourth speech synthesis method according to the present invention, information of a speech synthesis unit stored in advance includes at least a spectral parameter of a speech, and a formant emphasis filter in which a filter coefficient is determined according to the selected spectral parameter By shaping the formant of the signal and emphasizing the pitch of the synthesized speech signal by a pitch enhancement filter whose filter coefficient is determined according to the pitch parameter of the speech, a dull spectrum is shaped and at the same time the pitch harmonics are disturbed. It is intended to provide clear and high-quality synthesized speech.

According to a fifth aspect of the speech synthesis method of the present invention, the information of the speech synthesis unit stored in advance includes at least a speech spectrum parameter and a vocal tract filter drive signal of one pitch period, and the filter coefficient is selected according to the selected spectrum parameter. The formant enhancement filter that determines the pitch of the synthesized speech signal emphasizes the formant of the synthesized speech signal, and the pitch enhancement filter that determines the filter coefficient according to the pitch parameter of the speech enhances the pitch of the synthesized speech signal. The spectrum is shaped, and at the same time, clear and high quality synthesized speech without disturbance of pitch harmonics can be obtained.

In a sixth voice synthesizing method according to the present invention, information of a voice synthesizing unit stored in advance includes a waveform in which a formant of a waveform of at least one pitch period of a voice is emphasized, and a filter coefficient is set in accordance with a voice pitch parameter. By emphasizing the pitch of the synthesized speech signal by the determined pitch enhancement filter, it is possible to obtain high-quality synthesized speech that is clear and has no disturbance of pitch harmonics without performing formant enhancement processing during speech synthesis. It is a thing.

[0023]

BEST MODE FOR CARRYING OUT THE INVENTION

(First Embodiment) FIG. 1 shows the configuration of a voice synthesizing apparatus according to a first embodiment to which the first voice synthesizing method of the present invention is applied. This speech synthesizer includes a residual waveform storage unit 11, a voiced sound source generation unit 12, an unvoiced sound source generation unit 13, an LPC coefficient storage unit 14, an LPC coefficient interpolation unit 15, and a vocal tract filter unit 1.
6 and the formant emphasis filter unit 17 newly provided in the present invention.

The residual waveform storage unit 11 stores in advance a plurality of residual waveforms of one pitch cycle, which is the basis of the vocal tract filter drive signal, as information of a plurality of voice synthesis units. One 1-pitch cycle length residual waveform 102 selected according to the waveform selection information 101 is output.
The voiced sound source generation unit 12 repeats the one-pitch cycle length residual waveform 102 with the frame average pitch 103 as a cycle, and multiplies the repeated waveform by the frame average power 102 to generate the voiced sound source signal 105. This voiced sound source signal 105 is output in the voiced section discriminated by the voiced / unvoiced discrimination information 107, and the vocal tract filter unit 16
Is input to The unvoiced sound source generation unit 13 outputs an unvoiced sound source signal 106 represented by white noise or the like based on the frame average power 102. This unvoiced sound source signal 106 is
It is output in the unvoiced section determined by the voiced / unvoiced determination information 107 and input to the vocal tract filter unit 16.

The LPC coefficient storage unit 14 stores a plurality of LPC coefficients obtained as a result of performing linear prediction analysis (LPC analysis) on natural speech in advance as information on another plurality of speech synthesis units. One LPC coefficient 109 is selectively output according to the selection information 108. The residual waveform storage unit 11 extracts 1 from the residual waveform obtained by performing inverse filtering with these LPC coefficients.
The waveform of the pitch cycle is stored. LPC coefficient interpolator 1
5 is the LPC coefficient of the previous frame and the LPC coefficient of the current frame 1 so that the LPC coefficient does not become discontinuous between frames.
09 and outputs the LPC coefficient 110. The vocal tract filter unit 16 drives a vocal tract filter having an LPC coefficient 110 as a filter coefficient by the input voiced sound source signal 105 or unvoiced sound source signal 106, and synthesizes the synthesized speech signal 11
Outputs 1.

The formant emphasis filter unit 17
Filters the synthesized speech signal 111 with a filter coefficient determined according to the LPC coefficient 112 to emphasize the formants (the peaks of the spectrum), and
13 is output. That is, the formant emphasis filter requires a filter coefficient according to the spectrum parameter of the voice, but in this type of speech synthesizer, the filter coefficient of the vocal tract filter unit 16 is the spectrum parameter L.
Focusing on the setting according to the PC coefficient, the filter coefficient of the formant enhancement filter unit 17 is set according to the LPC coefficient 112 output from the LPC coefficient interpolating unit 15.

In this way, the formant emphasis filter unit 17
By emphasizing the formant of the synthesized speech signal 111, the blunted spectrum due to the causes described with reference to FIGS. 13 and 14 is shaped, and clear synthesized speech can be obtained.

FIG. 2 is a diagram showing another configuration example of the voiced sound source generation unit 12 different from the above. In the figure, the pitch cycle storage unit 24 stores the frame average pitch 103 and outputs the frame average pitch 204 of the previous frame. The pitch period interpolator 25 calculates the frame average pitch 204 of the previous frame from the frame average pitch 204 of the previous frame.
Then, the pitch cycle is interpolated so that the pitch cycle changes smoothly, and the waveform superposition position designation information 205 is output. The multiplier 21 multiplies the 1-pitch cycle length residual waveform 102 by the frame average power 102 to obtain the 1-pitch cycle length residual waveform 20.
Outputs 1. The pitch waveform storage unit 22 stores the 1-pitch cycle length residual waveform 201 and outputs the 1-pitch cycle-length residual waveform 202 one frame before. The waveform interpolating unit 23 weights the 1-pitch cycle length residual waveform 201 and the 1-pitch cycle length residual waveform 20 with weighting according to the waveform superposition position designation information 205.
2 interpolation is performed, and the 1-pitch cycle length residual waveform 20 after interpolation is performed.
3 is output. The waveform superimposition processing unit 26 generates and outputs the voiced sound source signal 105 by superimposing the one-pitch cycle length residual waveform 203 on the waveform superposition position designated by the waveform superposition position designation information 205.

Next, an example of the structure of the formant emphasis filter section 17 will be described. In the first configuration example, the formant enhancement filter is configured by an all-pole filter. The transfer function of this formant enhancement filter is expressed by the following equation.

[0030]

[Equation 1]

However, α _i is the LPC coefficient, N is the filter order, and β is a constant of 0 <β <1. Here, assuming that the transfer function of the vocal tract filter is H (z), Q ₁ (z) = H (z /
β), Q ₁ (z) is the pole p _i (i =
It can be said that 1, 1, ..., N) are replaced with β p _i (i = 1, ..., N). In other words, Q ₁ (z) is H
Since all the poles of (z) are brought close to the origin at a constant ratio β, the frequency spectrum of Q ₁ (z) becomes uneven as compared with H (z). Therefore, the larger β is, the larger the degree of formant enhancement is.

In the second configuration example of the formant emphasis filter section 17, the formant emphasis filter is constituted by a cascade connection of a pole-zero type filter and a first-order high-pass filter having a fixed characteristic. The transfer function of this formant enhancement filter is
It is expressed by the following equation.

[0033]

(Equation 2)

However, γ is a constant of 0 <γ <β, and μ is a constant of 0 <μ <1. In this case, the formant enhancement is performed by the pole-zero filter, and the extra spectral slope of the frequency characteristic of the pole-zero filter is corrected by the first-order high-pass filter.

The configuration of the formant enhancement filter unit 17 is not limited to the above two examples. Also,
A configuration in which the vocal tract filter section 16 and the formant emphasis filter section 17 are reversed in position is also possible. That is, since the vocal tract filter unit 16 and the formant emphasis filter 17 are both linear systems, the same effect can be obtained even if the positions are interchanged.

As described above, in the speech synthesizer of this embodiment, the formant emphasis emphasis filter unit 17 is arranged in cascade with the vocal tract filter unit 16 and the filter coefficient is set in accordance with the LPC coefficient. The spectrum of the synthetic speech signal which has been distorted due to the cause as described in 14 is shaped, and clear synthetic speech can be obtained.

(Second Embodiment) Next, FIG. 3 shows the configuration of a speech synthesizer according to a second embodiment of the present invention. FIG.
In FIG. 1, the components designated by the same reference numerals as those in FIG. 1 have the same functions as those in FIG.

In the present embodiment, the voiced / unvoiced discrimination information 10
In the unvoiced section determined by 7, the unvoiced sound source signal generated by the unvoiced sound source generation unit 13 is used as the drive signal, and the LPC coefficient 110 output from the LPC interpolation unit 15 is used as the filter coefficient in the unvoiced section. The unvoiced voice signal 303 synthesized by the vocal tract filter unit 16 is output. On the other hand, in the voiced section discriminated by the voiced / unvoiced discrimination information 107, the processing is performed by a procedure different from that in the first embodiment as described below.

The vocal tract filter unit 31 includes a residual waveform storage unit 1.
The 1-pitch cycle length residual waveform 102 output from 1 is used as a vocal tract filter drive signal, and the LPC coefficient 109 output from the LPC coefficient storage unit 14 is used as a filter coefficient to synthesize the 1-pitch cycle length speech waveform 301. The formant emphasis filter unit 17 converts the LPC coefficient 109 into the filter coefficient 112.
The 1-pitch cycle-long speech waveform 301 is filtered by the formant emphasis filter to emphasize the formants, and the 1-pitch cycle-long speech waveform 302 is output. The one-pitch cycle length speech waveform 302 is input to the voiced sound generation unit 32.

The voiced sound generator 32 can be realized with the same configuration as the voiced sound source generator 12 shown in FIG. However, when realizing the voiced sound generation unit 32 with the configuration of FIG.
In the voiced sound source generation unit 12, the one-pitch cycle length residual waveform 102
While 1 pitch cycle length speech waveform 30
Since 2 is input, the output is not the voiced sound source signal 105 but the voiced voice signal 304. Then, the unvoiced voice signal 303 is selected in the unvoiced section discriminated by the voiced / unvoiced discrimination information 107, and the voiced voice signal 304 is selected in the voiced section to output the synthesized speech signal 305.

According to this embodiment, when synthesizing a voiced speech signal, the filtering length in the vocal tract filter unit 31 and the formant emphasis filter unit 17 is 1
Since one pitch cycle length is sufficient for each frame and interpolation of the LPC coefficient is unnecessary, the same effect can be obtained with a smaller amount of calculation as compared with the first embodiment.

In this embodiment, the formant enhancement is performed only on the voiced voice signal, but the unvoiced voice signal 303 may also be provided with the formant enhancement filter unit to perform the formant enhancement as in the case of the voiced voice signal. Is.

Also in the present embodiment, the formant emphasis filter section 17 and the vocal tract filter section 31 may be arranged in opposite positions. (Third Embodiment) Next, FIG. 4 shows the configuration of a speech synthesizer according to a third embodiment of the present invention. 4, the components denoted by the same reference numerals as those in FIG. 3 have the same functions as those in FIG. 3, and the description thereof will be omitted.

In the second embodiment described with reference to FIG. 3, the formant enhancement is performed on the one-pitch cycle length speech waveform 301, whereas in this embodiment the synthesized speech signal 30 is used.
5 is different from the second embodiment in that the formant enhancement is performed on No. 5. Therefore, according to the present embodiment, the second
The same effect as that of the embodiment can be obtained.

(Fourth Embodiment) Next, FIG. 5 shows the configuration of a speech synthesizer according to a fourth embodiment of the present invention. FIG.
In FIG. 3, the components designated by the same reference numerals as those in FIG. 3 have the same functions as those in FIG.

In this embodiment, the 1-pitch cycle-long speech waveform is stored in the pitch-waveform storage unit 41, and the 1-pitch cycle-long speech waveform 302 is selected from the stored 1-pitch cycle-long speech waveform according to the waveform selection information 101. Output. here,
The one-pitch cycle length speech waveform stored in the pitch waveform storage unit 41 is a waveform in which the formant is emphasized by performing the processing shown in FIG. 6 in advance.

That is, in the present embodiment, the processing which was performed online in the configuration of FIG. 3 is performed offline in advance in the configuration of FIG. 6, and the residual waveform storage unit 11 and L are stored.
Residual waveform and LP output from the PC coefficient storage unit 14
Formant enhancement is performed on the synthesized speech signal 301 synthesized by the vocal tract filter unit 31 based on the C coefficient by the formant enhancement filter 112 to obtain a 1-pitch cycle-long speech waveform for all speech synthesis units, and these are calculated as pitch waveforms. The data is stored in the storage unit 41. Therefore, according to the present embodiment, it is possible to reduce the amount of calculation required for synthesizing the 1-pitch cycle length speech waveform and for enhancing the formant.

(Fifth Embodiment) FIG. 7 shows the arrangement of a speech synthesizer according to the fifth embodiment of the present invention. FIG.
In FIG. 5, the constituent elements denoted by the same reference numerals as those in FIG. 5 have the same functions as those in FIG. In this embodiment, the unvoiced voice 303 stored in the unvoiced voice storage unit 42 is output as the unvoiced voice 303 selected according to the unvoiced voice selection information 601. According to the present embodiment, compared with the fourth embodiment described with reference to FIG. 5, it is not necessary to perform filtering by the vocal tract filter when synthesizing the unvoiced voice signal, so that the calculation amount is further reduced.

(Sixth Embodiment) FIG. 8 shows the arrangement of a speech synthesizer according to the sixth embodiment of the present invention. FIG.
In FIG. 17, the components denoted by the same reference numerals as those in FIG. 17 have the same functions as those in FIG.

In this embodiment, a pitch emphasis filter section 51 is added to the structure of FIG. The pitch emphasis filter unit 51 is configured to adjust the frame average pitch 103
The synthesized speech signal 111 is filtered by the pitch emphasizing filter whose coefficient is determined in accordance with the above to emphasize the pitch and output the synthesized speech signal 501. The pitch emphasis filter unit 51 is realized by, for example, a filter having the following transfer function.

[0051]

(Equation 3) Here, p is the pitch period, and γ and λ are calculated according to the following equations based on the pitch gain.

[0052]

(Equation 4)

C _z and C _p are constants for controlling the degree of pitch enhancement and are determined empirically. Also, f
(X) is a control factor used to avoid unnecessary pitch enhancement when the signal to be processed is an unvoiced speech signal that does not include periodicity. x corresponds to the pitch gain, and when this x is smaller than a certain threshold value (typically 0.6), it is determined to be unvoiced sound and f (x) = 0. f when x is greater than or equal to the threshold
(X) = x. When x exceeds 1, f (x) = 1 is set in order to maintain stability. C _g is for absorbing the fluctuation of the gain of the filter between unvoiced sound and voiced sound, and is calculated by the following equation.

[0054]

(Equation 5)

According to the present embodiment, by newly providing the pitch emphasis filter section 51, in addition to the effect of clarifying the synthesized speech by shaping the blunted spectrum by formant emphasis in the above-described embodiments. As described above with reference to FIG. 15, the disturbance of the pitch harmonics of the synthesized speech signal based on the cause is improved, so that higher quality synthesized speech can be obtained.

(Seventh Embodiment) FIG. 9 shows the arrangement of a speech synthesizer according to the seventh embodiment of the present invention. In the present embodiment, the pitch enhancing filter unit 51 as described in the sixth embodiment is added to the speech synthesizer of the first embodiment described in FIG.

(Eighth Embodiment) Next, FIG. 10 shows the configuration of a speech synthesizer according to an eighth embodiment of the present invention. In FIG. 10, the components designated by the same reference numerals as those in FIG. 9 have the same functions as those in FIG.

In this embodiment, a gain adjusting section 61 is added to the speech synthesizer of the seventh embodiment described with reference to FIG. The gain adjusting section 61 includes a formant emphasis filter section 17 and a pitch emphasis filter section 51.
Is for correcting the total gain of, so that the powers of the final output synthetic voice signal 601 and the synthetic voice signal 111 output from the vocal tract filter unit 16 become equal to each other.
A multiplier 62 multiplies the output signal of the pitch enhancement filter section 51 by a predetermined gain.

(Ninth Embodiment) Next, FIG. 11 shows the configuration of a speech synthesizer according to a ninth embodiment of the present invention. In the present embodiment, a pitch enhancement filter section 51 is added to the speech synthesizer of the second embodiment described in FIG.

(Tenth Embodiment) Next, the first embodiment of the present invention will be described.
FIG. 12 shows the configuration of the speech synthesizer according to the No. 0 embodiment. In this embodiment, the voice synthesizer pitch enhancement filter unit 51 of the fifth embodiment described with reference to FIG. 5 is added.

[0061]

As described above, according to the present invention, a voice synthesizing method capable of generating a synthesized voice signal in which formant enhancement and further pitch enhancement are performed and thereby obtaining clear and high quality reproduced voice is provided. Can be provided.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a first embodiment according to the present invention.

FIG. 2 is a block diagram showing a configuration example of a voiced sound source generation unit according to the present invention.

FIG. 3 is a block diagram showing a second embodiment according to the present invention.

FIG. 4 is a block diagram showing a third embodiment according to the present invention.

FIG. 5 is a block diagram showing a fourth embodiment according to the present invention.

FIG. 6 is a block diagram showing an example of a method of generating a 1-pitch long speech waveform according to the present invention.

FIG. 7 is a block diagram showing a fifth embodiment according to the present invention.

FIG. 8 is a block diagram showing a sixth embodiment according to the present invention.

FIG. 9 is a block diagram showing a seventh embodiment according to the present invention.

FIG. 10 is a block diagram showing an eighth embodiment according to the present invention.

FIG. 11 is a block diagram showing a ninth embodiment according to the present invention.

FIG. 12 is a block diagram showing a tenth embodiment of the invention.

FIG. 13 is a diagram showing a relationship between a spectrum of a voice signal, a spectrum envelope, and a fundamental frequency.

FIG. 14 is a diagram showing a relationship between a spectrum of an analyzed speech signal and a spectrum of synthesized speech synthesized by changing a fundamental frequency.

FIG. 15 is a diagram showing the relationship between the frequency characteristics of two synthesis filters and the frequency characteristics of a filter obtained by interpolating them.

FIG. 16 is a diagram showing pitch disturbance of a voiced sound source signal.

FIG. 17 is a block diagram of a conventional speech synthesizer.

[Explanation of symbols]

 11 ... LPC coefficient interpolation section 12 ... Residual waveform storage section 13 ... LPC coefficient storage section 14 ... Unvoiced sound source generation section 15 ... Vocal tract filter section 16 ... Voiced sound source generation section 17 ... Formant enhancement filter section 41 ... Pitch waveform storage section 42 ... unvoiced voice storage unit 51 ... pitch enhancement filter unit 61 ... gain adjustment unit 62 ... multiplier

Claims (6)

[Claims] 1. A speech synthesis method for generating a synthesized speech signal by connecting information selected from a plurality of pieces of speech synthesis unit information stored in advance, wherein the speech synthesis unit information includes at least a spectrum parameter of speech. And a formant enhancement filter whose filter coefficient is determined according to the selected spectral parameter, to enhance the formant of the synthesized voice signal. 2. A speech synthesis method for generating a synthesized speech signal by connecting information selected from a plurality of pieces of speech synthesis unit information stored in advance, wherein the speech synthesis unit information is at least a speech spectrum parameter and 1 A method for synthesizing speech, which comprises enhancing a formant of the synthesized speech signal by a formant enhancement filter including a vocal tract filter driving signal having a pitch period and having a filter coefficient determined according to a selected spectral parameter. 3. A speech synthesis method for generating a synthesized speech signal by connecting information selected from a plurality of pieces of speech synthesis unit information stored in advance, wherein the speech synthesis unit information is at least one pitch period of speech. A method for synthesizing speech, including a waveform in which a formant of the waveform is emphasized. 4. A speech synthesis method for generating a synthesized speech signal by connecting information selected from a plurality of pieces of speech synthesis unit information stored in advance, wherein the speech synthesis unit information includes at least a spectrum parameter of speech. Shaping the formant of the synthesized speech signal by a formant enhancement filter whose filter coefficient is determined according to the selected spectral parameter, and changing the pitch of the synthesized speech signal by a pitch enhancement filter whose filter coefficient is determined according to the pitch parameter of the speech. A speech synthesis method characterized by emphasis. 5. A speech synthesis method for generating a synthesized speech signal by connecting information selected from a plurality of pieces of speech synthesis unit information stored in advance, wherein the speech synthesis unit information is at least a speech spectrum parameter and 1 A pitch in which a formant enhancement filter including a vocal tract filter driving signal having a pitch period and having a filter coefficient determined according to a selected spectral parameter emphasizes the formant of the synthesized speech signal and a filter coefficient is determined according to a pitch parameter of the speech. A voice synthesizing method, characterized in that a pitch of the synthesized voice signal is enhanced by an enhancement filter. 6. A speech synthesis method for generating a synthesized speech signal by connecting information selected from a plurality of pieces of speech synthesis unit information stored in advance, wherein the speech synthesis unit information is at least one pitch period of speech. A voice synthesizing method comprising a waveform emphasizing a formant of a waveform, further emphasizing a pitch of the synthesized voice signal by a pitch enhancement filter having a filter coefficient determined according to a voice pitch parameter.