JP3532059B2 - Speech synthesis method and speech synthesis device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To make it possible to set a pitch marl with little pitch fluctuation by a relatively simple processing and to achieve a high quality speech synthesis method by preparing speech waveform segments beforehand and winder/overlap- adding them with one pitch period shifted centering a minimum point in the speech waveform segments. SOLUTION: When a sentence containing kanji and hiragana letters is inputted to a text analysis part 101, the part executes morphemic analysis referring to a word dictionary 102, reading the sentence containing kanji and hiragana, determining accents and intonations, and outputting phonetic signs (intermediate language) with metrical sings. A parameter generation part 103 sets a pitch frequency pattern, a duration of phoneme, etc. A speech synthesis part 104 executes speech synthesis processing. Namely, it selects segments in a segment dictionary 105, windowing to hang the time-window of time-window length centering the pitch mark by a windowing part 106, and synthesizing speeches by PSOLA method (a synthesis method of superimposing speeches with pitch mark positions shifted in accordance with synthesis pitch period).

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice synthesizing method and a voice synthesizing apparatus for synthesizing an arbitrary voice according to a rule, and more particularly to a voice synthesizing method and a voice synthesizing apparatus for connecting voice waveforms to obtain a synthetic voice.

[0002]

2. Description of the Related Art As a conventional text-to-speech conversion apparatus, that is, a text-speech conversion apparatus for converting a text sentence into speech and outputting the speech, an apparatus composed of a text analysis section, a parameter generation section and a speech synthesis section is generally known. Has been. In the text analysis unit, a kanji / kana mixed sentence is input, morphological analysis is performed with reference to a word dictionary, pronunciation, accent, and intonation are determined, and a phonetic symbol with a prosody symbol (intermediate language) is output. In the parameter generator,
The pitch frequency pattern, phoneme duration, etc. are set. The voice synthesis unit performs a voice synthesis process. As a speech synthesis process in this speech synthesis unit, a linear prediction method or the like has been used before, but these methods result in deterioration of information. That is, since the vocal tract information and the sound source information, which are originally related to each other, are treated separately, and due to the restriction due to the modeling of the voice generation process, the deterioration of the sound quality cannot be avoided. For this reason, in recent years, a method of obtaining high-quality synthesized speech with little quality deterioration without separating vocal tract information and sound source information and using the original speech waveform as it is without artificial modeling has been used. Has become.

As a method of directly using a voice waveform, there is a conventional method: "FJ CHARPENTIER, MG STELL.
A, DIPHONE SYNTHESIS USING AN OVERLAP-ADD TECHNIQU
E FOR SPEECH WAVEFORMS CONCATENATION ”, Proc.Int.C
onf.ASSP, TOKYO, 1986 PP 2015-2018 ”is known. In this method, a pitch mark (reference point) is attached to the voice waveform in advance, the voice waveform is cut out around the position of the pitch mark, and the pitch mark position is overlapped by shifting the pitch mark position in accordance with the synthetic pitch period during synthesis. PSOLA (Pitch-Synchronous O
Verlap Add method).

FIG. 2 is cited from the above document and is a schematic diagram showing a PSOLA method for superimposing a voice waveform while changing a pitch. This schematic diagram shows an example in which the pitch period during synthesis is increased (pitch is lowered) as compared with the time of analysis (during segment production).

In the PSOLA method, since the pitch can be changed as needed, it has been widely used as a voice synthesizing unit in text-to-speech conversion. In this case, the pitch mark needs to be attached to a specific position for each pitch of the voice waveform, and the following positions have been proposed as the position of the pitch mark.

(1) For setting the peak of the voice waveform as the pitch mark setting position, for example, Japanese Patent Laid-Open No. 4-3729.
There is a "voice pitch conversion method" described in Japanese Patent Publication No. 99.

In this case, since the energy is concentrated at the local peak position of the voice waveform, it is considered to be suitable for storing the spectrum of the cut-out waveform.

(2) For setting the peak of the short time power as the setting position of the pitch mark, for example, "" Examination of speech synthesis system with waveform segment connection "Tsune Kawai, Yoshio Higuchi, Tohru Shimizu, Seiichi Yamamoto Technical report SP93-9 (1993-05)
The Institute of Electronics, Information and Communication Engineers.

Also in this case, as in the case of (1) above,
Since the energy is concentrated at the local peak position of the short-time power of the voice waveform, it is considered to be suitable for storing the spectrum of the cut-out waveform.

(3) A method in which the peak after the pitch filter is set as the pitch mark setting position is disclosed in, for example, Japanese Patent Application Laid-Open No. 7-
"Speech synthesis method and device" described in Japanese Patent No. 72897.
There is.

The peak after the pitch filter is the peak of the drive waveform of a 1-pitch vocal cord, and according to the above document, it is reported that it is a good representative of the pitch interval.

(4) As an example in which the 15% delay point of the impulse driving point is set as the pitch mark setting position, for example, "A study of pitch waveform extraction position" Yasuhiko Arai, Nishimura
Hiroshi Yoshida, Toshimitsu Minowa IEICE Technical Report SP95-8 (1995-0
5) The Institute of Electronics, Information and Communication Engineers.

According to this document, it is reported that the spectral distortion is minimized.

(5) As an example of setting the glottal closing point as the pitch mark setting position, for example, "About Japanese text-to-speech synthesis system using waveform superposition method" S. Sakamoto, T. Saito, K. Suzuki, Y. Hidehashi, Kobayashi There is Meishin Giho SP95-6 (1995-05) The Institute of Electronics, Information and Communication Engineers.

The glottal closing point in this document is considered to be the same as the impulse driving point (excitation point of one pitch waveform). In order to stably extract this glottal closing point,
Dynamic Wavelet transform is used.

[0016]

However, the conventional pitch mark position as described above has the following problems.

According to the experiment of the applicant of the present invention, there is a gap between the peak position of the waveform and the peak position after the pitch filter of the above (3), and according to the experiment by the applicant, the pitch caused by the gap is changed. The shaking is great and the sound becomes muffled. Relatively good results were obtained when the peak of the voice waveform of (1) was set as the pitch mark setting position.
In the case of (2) in which the power peak for a short time is set as the set position of the pitch mark, the maximum value and the minimum value are evaluated equally, so that pitching may occur depending on the speaker. In the case where the pitch mark setting position is the 15% delay point of the impulse driving point in (4), the processing amount for specifying the setting position is large, resulting in a delay in processing, and depending on the individual and the type of phoneme. , The delay point of 15% is not always the best. In the case where the glottal closure point in (5) is set as the pitch mark setting position, this is performed to extract this glottal closure point.
The dynamic wavelet conversion requires a large amount of processing and causes a delay in processing as in (4) above.

The present invention has been made in view of the above problems, and realizes a high-quality voice synthesizing method and voice synthesizing apparatus by enabling pitch mark setting with less pitch fluctuation with relatively simple processing. The purpose is to do.

[0019]

In order to solve the above-mentioned problems, a speech synthesis method according to the first invention is directed to a step of detecting a local minimum point immediately before a peak of an audio signal, and a detected local minimum point. A voice synthesis unit is created in advance by the step of centering and cutting out the voice signal, and window overlapping is performed while shifting a pitch period with a minimum point in the voice synthesis unit as a center of superimposition. To do.

As described above, since the minimum point immediately before the peak of the audio signal is set as the center point of the audio signal to be cut out, the center point for superimposing can be easily set by a simple process, and the spectral distortion Can also be smaller. As a result, the rumbling sound was reduced.

[0021] Further, by treating a speech unit in units of a fixed length and performing frame processing, during speech synthesis,
It becomes easier to control the voice waveform data.

In the speech synthesis method according to the second aspect of the invention, the step of detecting the pitch period of the speech signal, the step of detecting the minimum point immediately before the peak of the speech signal, and the centering of the detected minimum point are performed. A voice synthesis unit is created in advance by a step of cutting out the voice signal and a step of multiplying the cut out voice signal by a window that is a constant multiple of the pitch period, and a minimum point in the voice synthesis unit is superimposed. It is characterized in that they are overlapped while being shifted by a pitch period as a center.

As described above, the elemental pieces are windowed in advance based on the obtained pitch period, so that it is not necessary to perform windowing processing during voice synthesis. As a result, the processing amount at the time of speech synthesis processing can be significantly reduced, and the processing device can be simplified or the processing speed can be increased.

In the voice synthesizing method according to the third aspect of the invention, the steps of appropriately inverting the positive and negative signs of the voice signal to match the positive and negative signs of the whole voice signal, and the step of detecting the minimum point immediately before the peak of the voice signal are detected. A voice synthesis unit is created in advance by the step of centering the minimum point and cutting out the voice signal, and windowing is performed while shifting the minimum point in the voice synthesis unit as the center of superimposition for a pitch period. It is characterized by overlapping.

With the above configuration, it is possible to digitally correct a change in phase due to a change in the configuration of the analog system.

A voice synthesizing method according to a fourth aspect of the present invention includes a step of detecting a pitch period of a voice signal and a step of appropriately inverting the positive and negative of the voice signal to match the positive and negative of the entire voice signal.
A step of detecting a local minimum point just before the peak of the audio signal; a step of centering the detected local minimum point to cut out the audio signal; and a step of multiplying the cut out audio signal by a window that is a constant multiple of the pitch period. And a voice synthesis unit is created in advance, and the voice synthesis unit is superimposed while shifting the minimum point in the voice synthesis unit by the pitch period.

With the above configuration, since the element is windowed in advance, the processing amount at the time of speech synthesis processing can be greatly reduced.

Further, since the function of inverting the positive / negative of the voice waveform is provided, it is possible to digitally correct the phase change due to the change of the configuration of the analog system.

A voice synthesizing apparatus according to a fifth aspect of the present invention comprises a minimum point detecting means for detecting a minimum point immediately before a peak of a voice signal,
A voice signal cutout unit for centering the minimum point detected by the minimum point detection unit to cut out the voice signal, and a voice synthesis unit for storing the voice synthesis unit cut out by the voice signal cutout unit. The present invention is characterized by comprising a storage means and a voice synthesis section for window-applying the voice synthesis element stored in the voice synthesis element storage means while shifting the minimum point of the voice synthesis element by a pitch period and superimposing it. .

As described above, since the local minimum point immediately before the peak of the audio signal detected by the local minimum point detecting means is set as the center point of the audio signal cut out by the audio signal cutting out means, the center when superimposing in the audio synthesizing section. Points can be easily set by simple processing, and spectral distortion can be reduced. As a result, the rumbling sound was reduced. In addition, by treating the speech units in units of a fixed length and performing frame processing, it becomes easier to control the speech waveform data during speech synthesis.

A voice synthesizer according to a sixth aspect of the invention is a pitch period detecting means for detecting a pitch period of a voice signal, a minimum point detecting means for detecting a minimum point immediately before a peak of a voice signal, and the minimum point detecting means. An audio signal cutting-out means for cutting out the audio signal by centering on the minimum point detected by, and a windowing means for multiplying the audio signal cut out by the audio signal cutting-out means with a window which is a constant multiple of the pitch period. A voice synthesis unit storage unit that stores the voice synthesis unit that has been windowed by the window unit, and the voice synthesis unit stored in the voice synthesis unit storage unit with the minimum point being the center of superimposition. , And a voice synthesizing unit for superimposing while shifting by a pitch period.

As described above, the windows are preliminarily windowed by the windowing means on the basis of the pitch period obtained by the pitch period detecting means, so that it is not necessary to perform the windowing process at the time of speech synthesis. As a result, the processing amount at the time of speech synthesis processing can be significantly reduced, and the processing device can be simplified or the processing speed can be increased.

The voice synthesizer according to the seventh aspect of the invention comprises a voice signal inverting means for appropriately inverting the positive and negative of the voice signal to match the positive and negative of the whole voice signal, and a minimum inspection for detecting a minimum point immediately before the peak of the voice signal. The output means, the voice signal cutout means for centering the minimum point detected by the minimum point detection means to cut out the voice signal, and the voice synthesis unit cut out by the voice signal cutout means are stored. A voice synthesis unit storage means and a voice synthesis unit for windowing and superimposing the voice synthesis unit stored in the voice synthesis unit storage unit while shifting by a pitch period with the minimum point as a center of superposition. Is characterized by.

With the above-described structure, the positive and negative signs of the audio signal are appropriately inverted by the audio signal inverting means to match the positive and negative signs of the entire audio signal, thereby digitally correcting a phase change due to a change in the structure of the analog system. be able to.

A voice synthesizer according to an eighth aspect of the present invention comprises a pitch period detecting means for detecting a pitch period of a voice signal, and a voice signal inverting means for appropriately inverting the positive / negative of the voice signal to match the positive / negative of the entire voice signal. A local minimum point detecting means for detecting a local minimum point immediately before the peak of the audio signal; an audio signal clipping means for centering the local minimum point detected by the local minimum point detecting means for clipping the audio signal; Windowing means for applying a window having a constant multiple of the pitch period to the voice signal cut out by the means, voice synthesis element storage means for storing the voice synthesis element windowed by the windowing means, And a voice synthesis unit for superimposing the voice synthesis unit stored in the voice synthesis unit storage unit while shifting the minimum point of the voice synthesis unit by a pitch period.

With the above construction, the element is pre-windowed by the windowing means, so that the processing amount at the time of speech synthesis processing can be greatly reduced.

Further, since the audio signal inverting means for appropriately inverting the positive and negative of the audio waveform to match the positive and negative of the entire audio signal is provided, it is possible to digitally correct the phase change due to the change of the configuration of the analog system. it can.

A speech synthesizer according to a ninth aspect of the invention comprises an inverting amplifier for inverting and amplifying a voice signal, a non-inverting amplifier for non-inverting amplification of the voice signal, a voice signal from the inverting amplifier and a non-inverting amplifier. A selector that selects an audio signal, an AD converter that converts the audio signal selected by the selector into a digital value, a storage unit that stores the data AD-converted by the AD converter, and a storage unit that stores the data. The audio signal reading means for sequentially reading the audio signals, the minimum point detecting means for detecting the minimum point immediately before the peak of the audio signal read by the audio signal reading means, and the minimum point detected by the minimum point detecting means. A voice signal cutout unit for centering the voice signal by centering and a voice synthesis unit cut out by the voice signal cutout unit are stored. And a voice synthesis unit storage unit, and a voice unit connection synthesis unit for windowing and superimposing while shifting a pitch period with a minimum point of the voice synthesis unit selected from the voice synthesis unit storage unit as a center of superposition. It is characterized by that.

With the above configuration, the inverting amplifier or the non-inverting amplifier and the selector appropriately select the inverting-amplified audio signal and the non-inverting-amplified audio signal, and appropriately invert the positive and negative of the audio waveform. Match the positive and negative of the entire signal.
As a result, it is possible to digitally correct a change in the phase due to a change in the configuration of the analog system.

Further, since the local minimum point immediately before the peak of the audio signal detected by the local minimum point detecting means is used as the center point of the audio signal cut out by the audio signal cutting means, the center point at the time of superimposing can be easily performed by a simple process. Can be set to
Spectral distortion can also be reduced. As a result,
The rumbling sound decreased. Also, by handling speech units with a fixed length as a unit and performing frame processing,
It becomes easy to control the voice waveform data during voice synthesis.

The speech synthesizer according to the tenth aspect of the invention comprises an inverting amplifier for inverting and amplifying a voice signal, a non-inverting amplifier for non-inverting amplification of the voice signal, a voice signal from the inverting amplifier and a non-inverting amplifier. A selector for selecting an audio signal, an AD converter for converting the audio signal selected by the selector into a digital value, and an AD for the AD converter.
Storage means for storing the converted data, audio signal reading means for sequentially reading the audio signals stored in the storage means, and pitch cycle detecting means for detecting the pitch cycle of the audio signal read by the audio signal reading means. Window length calculation means for multiplying the pitch cycle detected by the pitch cycle detection means by a constant, minimum point detection means for detecting a minimum point immediately before the peak of the audio signal read by the audio signal reading means, and the minimum point detection. An audio signal cutout means for centering the minimum point detected by the means to cut out the audio signal, and a windowing of the window length calculated by the window length calculation part on the audio signal cut out by the audio signal cutout means. Windowing means, voice synthesis element storage means for storing the voice synthesis element that has been windowed by the windowing section, and the voice synthesis element description. As the center of superimposing the minimum point of the speech synthesis fragments selected from among means, characterized in that a segment connection combining unit that superimposes shifting pitch cycle.

With the above configuration, a phase change due to a change in the analog system configuration or the like can be digitally corrected, and the windowing means can preliminarily use the element pieces based on the pitch period obtained by the pitch period detecting means. Since it is windowed, it is not necessary to perform windowing processing during voice synthesis. As a result, the processing amount at the time of speech synthesis processing can be significantly reduced, and the processing device can be simplified or the processing speed can be increased.

[0043]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described below with reference to the accompanying drawings.

[First Embodiment] A voice synthesizing method and a voice synthesizing apparatus according to the first embodiment will be described below. FIG. 1 is a block diagram showing the configuration of the speech synthesizer according to the first embodiment.

In the text analysis unit 101, when a kanji / kana mixed sentence is input, morphological analysis is performed with reference to the word dictionary 102 to determine readings, accents and intonations of the kanji / kana mixed sentence, and phonetic symbols with prosodic symbols. Output (intermediate language). The parameter generator 103 sets a pitch frequency pattern, phoneme duration, and the like. The text analysis unit 101, the word dictionary 102, and the parameter generation unit 103 are the same as the conventional ones.

The voice synthesis unit 104 performs a voice synthesis process. That is, a segment in the segment dictionary 105 is selected, and a windowing unit 106 performs windowing for multiplying the segment by a time window having a time window length Tp1 described later so that the pitch mark is at the center, and P
Speech synthesis is performed by the SOLA method.

Here, the time window length Tp1 is set as Tp1 = C0 × min (Tpa, Tps), where Tpa is the pitch period during analysis and Tps is the pitch period during synthesis. Note that C0 has a value of about 2.0.

The segment dictionary 105 is a dictionary in which segments are written. The segment is created by the segment creating unit 107. The segment creating unit 107 is a main part of the present invention and has a processing function shown in the flowchart of FIG.

The processing in the segment creating unit 107 will be described with reference to FIG. When an audio signal is input to the segment creating unit 107 by the audio signal input unit 108 including a data disc, first, in step S201, the input audio signal data is divided into analysis frames. This analysis frame is audio signal data divided into sections of a fixed length. In this embodiment, one frame length is 32 msec and 8 m
It is delimited so that it shifts to the next frame and moves to the next frame.
Here, the total number of frames is N.

In step S202, the frame number i of the analysis frame to be processed is initialized. Step S20
In 3, the coordinate xd on the time axis that gives the minimum value immediately before the peak in the i-th frame is detected. An example of detecting this coordinate xd is shown in FIG. The voice waveform shown in FIG. 11 is the voice waveform when uttered as a (/ a /), and the mark "*"
Is the pitch mark position (the minimum value immediately before the peak) according to the present embodiment. This pitch mark position can be easily detected. That is, the peak point in each analysis frame can be easily specified, and the local minimum point immediately before that can also be easily specified. Since this minimum point is the pitch mark position, the pitch mark position can be easily detected.

Then, in step S204 in FIG. 3, L audio data is cut out before and after the coordinate xd and centered so that the coordinate xd is located at the center. In addition,
Here, L minute was set to 12 msec. This is a value in which a margin is given to the longest pitch period in males by a preliminary experiment by the present inventor.

In step S205, the speech data extracted in step S204 is stored in a storage medium such as a data disk as a segment in the i-th frame and stored in the segment dictionary 105.
Are sequentially written. In step S206, it is determined whether writing of the segment has been completed for all the analysis frames. If this writing has not been completed, the frame number is updated in step S207 and then step S2 is performed.
Returning to step 03, the processing from step S203 to step S205 is continued. If it is determined in step S206 that all analysis frames have been processed, the segment dictionary 105
The data disc closing process (not shown) and the like are performed, and the operation of the segment creating unit 107 ends.

The target segment is appropriately selected from the segment dictionary 105 in which the segment created by the above processing is written, and windowing is performed by the windowing unit 106, and the voice synthesis unit 104 is selected. The voice synthesis process is performed.

It is assumed that the local minimum value detection just before the peak is performed only on the voiced part of the voice signal. The voice data is used as it is for the unvoiced part. The same applies to other embodiments described below.

[Effect] Since the minimum value immediately before the peak in each analysis frame is used as the pitch mark, the pitch mark can be set by a simple process and the spectral distortion can be reduced. As a result, according to the experiments by the present inventor, the sound that was rumbling in hearing was reduced.

Further, since the voice segment is handled in units of a fixed length of 12 msec and the frame processing is performed, there is an effect that voice waveform data can be easily controlled during voice synthesis.

[Second Embodiment] Next, a second embodiment of the present invention will be described.

FIG. 4 is a block diagram showing the configuration of the speech synthesizer according to the second embodiment.

In the speech synthesizer according to this embodiment,
The text analysis unit 101, the word dictionary 102, the parameter generation unit 103, and the voice signal input unit 108 are the same as those of the voice synthesis device according to the first embodiment.

The voice synthesizing method of this embodiment is similar to the voice synthesizing method of the first embodiment in that the pitch mark is set to the minimum point immediately before the peak in each analysis frame. The feature of the speech synthesis method of the present embodiment is that the segment creation unit 301 performs windowing on the segment in advance.

The segment creating section 301 is composed of a pitch mark calculating section 302 and a windowing section 303. The segment creating unit 301 has the processing function shown in FIG.
The processing in the segment creating unit 301 will be described below.

When the audio signal data is input from the audio signal input unit 108, the audio signal data is first divided into analysis frames in step S401. This analysis frame
Similar to the first embodiment, one frame length is 32 msec, and it is set to shift to the next frame by shifting 8 msec. The total number of frames is N.

In step S402, the frame number i for processing is initialized. In step S403, the pitch period Tp of the voice in the i-th frame is detected. As a method of detecting the pitch period Tp, a method of detecting the peak interval of the waveform can be considered as a simple method, but in the present embodiment, the cepstrum method is used. This is for more accurately calculating the pitch period. In this cepstrum method, the pitch cycle Tp is detected in the processing steps shown in FIG. First, a time waveform is input in step S501, and windowing is performed in step S502. Next, the windowed time waveform is subjected to discrete Fourier transform (DFT) in step S503, and the square root of the sum of squares of its real part and imaginary part is logarithmically converted in step S504. After that, step S
Inverse Fourier transform (IDFT) is performed in 505, and the cepstrum component is obtained and output in step S506. As described above, the cepstrum method converts a convolution operation into an additive operation. Since the voiced sound signal of speech is a convolution of the sound source component with vocal tract information, the cepstrum method is suitable for separating the two. When the input signal is a voiced voice signal, if the pitch period is T0, the sound source component appears near T0 of high Keflenshiy (long-term region), and the vocal tract component is a low Keflenshiy (short-term region) component. appear.
In order to obtain the pitch period from the cepstrum, the peak of the high kefrenshi part can be obtained and the time from the time origin to this peak can be measured.

Next, in step S404 in FIG. 5, the i-th
The coordinate xd on the time axis that gives the minimum value immediately before the peak in the frame is detected. The detection of the coordinate xd is the same as that in the first embodiment (see FIG. 11). In step S405, audio data for L before and after the coordinate xd is cut out and centered so that the coordinate xd is located at the center. The length of L is the same as that in the first embodiment.

Next, in step S406, the pitch period Tp obtained in step S403 is multiplied by a constant C1,
In step S407, the front and rear are centered around the minimum point xd, and C1 ×
Multiply a time window of length Tp. In this embodiment, a value of about 1.0 is used as the constant C1. According to the experiments of the present inventor, a value smaller than 1.0 is desirable as the constant C1. This is because when the constant C1 is smaller than 1.0, it is possible to suppress the influence of adjacent pitches and reduce noise.

Then, in step S408, the windowed audio data is sequentially written as a segment in the i-th frame to a storage medium such as a data disk as a segment dictionary 305. Next, in step S409, it is determined whether or not the processing for all the frames has been completed. If not completed, the frame number is updated in step S410, the process returns to step S403, and the processing from step S403 to step S408 is performed. continue. When it is determined in step S409 that the processing of all the frames has been completed, the processing of closing the storage medium such as the data disk (not shown) or the like is performed and the operation of the segment creating unit 301 is completed.

The target segment is appropriately selected from the segment dictionary 305 in which the segment created by the above process is written, and the voice synthesis unit 304 performs the voice synthesis process.

[Effect] In the present embodiment, the segment creating unit 30.
In 1, the pitch period Tp is accurately obtained, and the element to be written in the element dictionary 305 is windowed in advance based on the obtained value. Therefore, it is not necessary to perform the windowing process at the time of speech synthesis.
That is, the windowing process (multiplication) for each pitch, which is required in the voice synthesis process in the first embodiment, is unnecessary,
Since it is only necessary to execute the superposition, it is possible to significantly reduce the processing amount at the time of speech synthesis processing.

As a result, the voice synthesizing apparatus using the voice synthesizing method of the present embodiment can be realized by a normal CPU without using an advanced arithmetic processor such as DSP. Further, when the same arithmetic processor is used, it is possible to significantly speed up the voice synthesis process.

By setting the constant C1 to a value smaller than 1.0, it is possible to suppress the influence of adjacent pitches and reduce noise.

[Third Embodiment] Next, a third embodiment of the present invention will be described.

The overall structure of the speech synthesizing apparatus used in the speech synthesizing method of this embodiment is almost the same as that of the speech synthesizing apparatus of the first embodiment. The feature of the present embodiment is that the function of controlling the phase inversion of the voice waveform is provided in the processing in the segment creating unit (107), and the inversion control unit is provided, and the phase is created in the segment creating unit. The point is that it is possible to control whether or not to perform the inversion process.

First, the operation of the segment creating section will be described with reference to the flowchart of FIG.

When the voice signal data is input, the voice signal data is divided into analysis frames in step S601.
In step S602, the frame number i to be processed is initialized. These processes are the same as those in the first embodiment.

In step S603, the inversion flag stored in the shared memory is checked. If the inversion flag is 1, the positive / negative of the voice waveform is inverted in step S604. If the inversion flag is 0, the voice waveform is not inverted, and the process jumps to step S605.

Next, in step S605, the coordinate xd on the time axis that gives the minimum value immediately before the peak in the i-th frame.
To detect. The detection of the coordinate xd is the same as that in the first embodiment (see FIG. 11). Step S606
Then, L audio data before and after the coordinate xd is cut out and centered so that the coordinate xd is located at the center.
The length of L is the same as that in the first embodiment. Next, in step S607, the voice data cut out in step S606 is sequentially written in the segment dictionary 105 as a segment in the i-th frame.

Then, in step S608, it is determined whether or not the processing of all the frames has been completed. If not completed, the frame number is updated in step S609 and the process returns to step S603. From step S603 to step S607 Continue processing. When it is determined in step S608 that the processing of all the frames has been completed, the processing of closing the storage medium such as the data disk (not shown)
Then, the operation of the segment creating unit is finished.

The target speech segment is appropriately selected from the speech segment dictionary 105 in which the speech segment created by the above processing is written, and the speech synthesis unit 104 performs speech synthesis processing.

Next, the operation of the inversion controller will be described. The reversing control unit controls the reversing process of the voice waveform in the segment creating unit based on an instruction from the operator from a keyboard or the like, and has a processing function shown in the flowchart of FIG. 7B. ing. The operation of this inversion control unit will be described below.

First, in step S610, the operator's intention input from a keyboard or the like is confirmed. That is, it is determined whether or not the operator has instructed to invert the phase of the audio signal. In the case of the inversion instruction, the inversion flag in the shared memory (the shared memory checked in step S603 of the segment creating unit) in the shared memory is set to 1 in step S611. In the case of the non-inversion instruction, the inversion flag on the shared memory is set to 0 in step S612.

Based on the setting of the inversion flag on the shared memory, the determination in step S603 of the segment creating unit is made.

It should be noted that it is desirable to execute the inversion control section first if the analog system is constant. If the environment in which the audio signal is recorded is partially different from the others,
The audio signal input unit 108 of FIG. 1 and the inversion flag set on the shared memory may be associated with each other to create a table, and the inversion flag may be stored in the shared memory based on the table.

[Effect] When an analog system such as a microphone or an amplifier for amplifying a voice signal picked up by the microphone is changed,
If the analog system differs from the original configuration, the phase may be inverted. In this case, the positive and negative signs of the audio data are reversed, so the intention is to detect the minimum value, but the maximum value may be detected.

According to this embodiment, it becomes possible to digitally correct the phase change due to the change in the configuration of the analog system. As a result, a single voice synthesizer can handle the difference in analog system.

[Fourth Embodiment] Next, a fourth embodiment of the present invention will be described.

The overall configuration of the speech synthesizer of this embodiment is as follows.
It is almost the same as the speech synthesizer according to the second and third embodiments. The features of the present embodiment in comparison with the second embodiment are that the function of controlling the phase inversion of the voice waveform is provided in the processing of the segment creating unit, and that the segment creating unit is provided with an inversion control unit. The point is that it is possible to control whether or not the phase inversion process is performed. The feature of the present embodiment in comparison with the third embodiment is that the elemental piece creating unit performs windowing on the elemental piece in advance.

First, the operation of the segment creating section will be described with reference to the flowchart of FIG.

When the voice signal data is input, the voice signal data is divided into analysis frames in step S701,
In step S702, the frame number i to be processed is initialized. These processes are the same as those in the first embodiment. In step S703, the pitch period Tp of the voice in the i-th frame is detected. As a method for detecting the pitch period Tp, the cepstrum method is used as in the second embodiment.

In step S704, the inversion flag stored in the shared memory is checked. If the inversion flag is 1, the positive / negative of the voice waveform is inverted in step S705. If the inversion flag is 0, the voice waveform is not inverted, and the process jumps to step S706.

Next, in step S706, the coordinate xd on the time axis which gives the minimum value immediately before the peak in the i-th frame.
To detect. The detection of the coordinate xd is the same as that in the first embodiment (see FIG. 11). Step S707
Then, L audio data before and after the coordinate xd is cut out and centered so that the coordinate xd is located at the center.
The length of L is the same as that in the first embodiment. Then, in step S708, the step S703 is performed.
The pitch period Tp obtained in step S1 is multiplied by a constant C1, and step S709
Then, a time window with a length of C1 × Tp is applied to the front and back around the minimum point xd.

Then, in step S710, the windowed audio data is sequentially written as a segment in the i-th frame to a storage medium such as a data disk as a segment dictionary (305). Next, in step S711, it is determined whether or not the processing for all the frames has been completed. If not completed, the frame number is updated in step S712, the process returns to step S703, and the processes from step S703 to step S710 are performed. continue. Step S71
When it is determined in 1 that the processing of all the frames is completed, the processing of closing the storage medium such as the data disk (not shown) is performed and the operation of the segment creating unit is completed.

The target segment is appropriately selected from the segment dictionary in which the segment created by the above process is written, and the speech synthesis unit performs the speech synthesis process.

Next, the operation of the inversion controller will be described. This inversion control unit is similar to the inversion control unit in the third embodiment and has the processing function shown in the flowchart of FIG. 8 (B). The operation of this inversion control unit will be described below.

First, in step S720, the operator's intention input from the keyboard or the like is confirmed. That is, it is determined whether or not the operator has instructed to invert the phase of the audio signal. In the case of the inversion instruction, the inversion flag on the shared memory (the shared memory checked in step S704 of the segment creating unit) in the shared memory is set to 1 in step S721. In the case of the non-inversion instruction, the inversion flag on the shared memory is set to 0 in step S722. Based on the setting of the inversion flag on the shared memory, the determination in step S704 of the segment creating unit is made.

The inversion control unit is preferably executed first if the analog system is constant, as in the case of the inversion control unit in the third embodiment. If the environment in which the audio signal is recorded is partially different from the others, a table is created by associating the audio signal input section with the inversion flag set on the shared memory, and the inversion flag is set based on this table. It may be configured to be stored in the shared memory.

[Effect] According to the fourth embodiment, since the pitch cycle detecting unit is provided in the segment creating unit and the segment to be written in the segment dictionary is windowed in advance, it is windowed at the time of speech synthesis. Eliminates the need for processing. That is, the windowing process (multiplication) for each pitch becomes unnecessary, and the processing amount at the time of voice synthesis processing can be significantly reduced.

Further, since the function of inverting the positive / negative of the voice waveform is provided, it becomes possible to digitally correct the phase change due to the change of the configuration of the analog system. As a result, a single voice synthesizer can handle the difference in analog system.

[Fifth Embodiment] Next, a fifth embodiment of the present invention will be described. FIG. 9 shows the configuration of a speech synthesizer according to the fifth embodiment.

The audio signal input to the audio input terminal 800 in the figure is input to the inverting amplifier 802 as the path 1 and the phase thereof is inverted, and then input to the selector 803. The other path is input to the selector 803 via the non-inverting amplifier 801 (without inverting the phase of the audio signal). In the selector 803, one of the audio signal passed through the inverting amplifier 802 and the audio signal passed through the non-inverting amplifier 801 is selected and input to the AD converter 804. The input audio signal is converted into a digital signal by the AD converter 804 and stored in the storage medium 805.

The audio signal reading circuit 806 reads the audio data stored in the storage medium 805, and the local minimum value detecting circuit 807 detects the local minimum value immediately before the peak. The minimum value detection processing by the minimum value detection circuit 807 is performed in step S2 of the segment creating unit 107 in the first embodiment.
This is the same as the minimum value detection processing in 03 (see FIG. 3) (see FIG. 11).

The voice cut-out circuit 808 cuts out L amount of voice data before and after the local minimum value immediately before the peak.
Center so that this minimum value is located in the center. L
The minute was set to 12 msec as in the first embodiment. The voice cut-out processing in the voice cut-out circuit 808 is performed in step S of the segment creating unit 107 in the first embodiment.
This is the same as the audio cutout process in 204 (see FIG. 3). The data cut out by the audio cutout circuit 808 is stored in a disk device or the like as a segment file 809.

The above processing operation is performed for all the input voice data.

Next, the operation at the time of speech synthesis processing will be described.

When a character string is input to the character phoneme symbol conversion circuit 811 via the character string input terminal 810, the character phoneme symbol conversion circuit 811 adds a corresponding accent mark to the input character string. Output phoneme symbols. In the prosody information setting circuit 812, the intonation strength of the phoneme symbol from the character phoneme symbol conversion circuit 811,
Prosody information such as phoneme duration is set.

In the segment selection circuit 813, the segment file 8 is used to extract the segment necessary for converting the phoneme symbol string into speech.
It is selected from 09, read out, and output to the windowing circuit 814. In this windowing circuit 814, the segment selection circuit 813
Windowing is performed for each frame of the segment read out in (4) and output to the segment connection synthesis circuit 815. Element connection synthesis circuit 8
At 15, the pieces of each frame that have been windowed by the windowing circuit 814 are shifted by the combined pitch period and overlapped. By the above operation during voice synthesis, a time waveform of voice is obtained and output from the synthesized voice output terminal 816.

[Effect] The inverting amplifier 802, the non-inverting amplifier 801 and the selector 803 even when the analog system such as a microphone or an amplifier is changed and the phase of the audio signal is inverted unlike the original configuration. Thus, the change in the phase can be easily corrected in an analog manner. As a result, a single voice synthesizer can cope with the phase difference of the analog system.

[Sixth Embodiment] Next, a sixth embodiment of the present invention will be described. FIG. 10 shows the configuration of a speech synthesizer according to the sixth embodiment. Since the overall configuration of the speech synthesizer according to the present embodiment is almost the same as that of the speech synthesizer according to the fifth embodiment, the same parts will be denoted by the same reference numerals.

The voice signal input to the voice input terminal 900 in the figure is the same as in the fifth embodiment, except that the inverting amplifier 9
The signal is input to the selector 903 through the path through 02 and the path through the non-inverting amplifier 901. Selector 903
Then, one of the audio signal passing through the inverting amplifier 902 and the audio signal passing through the non-inverting amplifier 901 is selected and A
It is input to the D converter 904. The input audio signal is
This AD converter 904 converts it into a digital signal and stores it in a storage medium 905.

The audio signal reading circuit 906 reads the audio data stored in the storage medium 905 and outputs it to the pitch cycle detecting circuit 921. Pitch cycle detection circuit 9
At 21, the pitch period of the voice data is detected. This pitch cycle detection processing is performed by the segment creating unit 3 of the second embodiment.
This is the same as the 01 pitch period detection processing (step S403). A cepstrum method or the like is used as the pitch period detection method. After the detection processing in the pitch period detection circuit 921,
The minimum value detection circuit 907 detects the minimum value immediately before the peak.

The voice cut-out circuit 908 cuts out voice data of L minutes (12 msec as in the first embodiment) before and after the local minimum value immediately before the peak, and centers the local minimum value. . The data cut out by the audio cutout circuit 908 is the windowing circuit 92.
3 is output. The windowing circuit 903 multiplies the time window having the time window length calculated by the window length calculation circuit 922 based on the pitch cycle from the pitch cycle detection circuit 921. The windowed audio data is stored in a disk device or the like as a fragment file 909.

The above processing operation is performed for all the input audio data.

Next, the operation at the time of voice synthesis will be described.

With respect to the character string input to the character string input terminal 910, the character phoneme symbol conversion circuit 911 outputs a corresponding phoneme symbol with an accent symbol. The prosody information setting circuit 912 sets prosody information such as the intensity of intonation and the duration of the phoneme in this phoneme symbol.

In the segment selection circuit 913, the segment necessary for converting the phoneme symbol string into speech is stored in the segment file 9
It is selected from 09 and read out, and it outputs to the element connection synthesis circuit 915.

In the unit piece connection synthesizing circuit 915, the unit pieces are superimposed for each frame while being shifted by the synthetic pitch period. By the above-described operation during voice synthesis, a time waveform of voice is obtained and output from the synthesized voice output terminal 916.

[Effect] According to the present embodiment, since windowing is performed when the voice unit file is created, windowing is not required during voice synthesis. Therefore, the circuit configuration of the voice synthesizing portion becomes simple without including a multiplier.

Further, by providing the inverting amplifier and the selector, the inversion of the phase of the voice data can be corrected in an analog manner and easily, so that the same voice synthesizer can be applied. .

[Modification] Although the cepstrum method is used as the pitch period detection method in the second, fourth, and sixth embodiments, other methods such as the autocorrelation method and the autocorrelation of linear prediction residuals are used. Other methods such as the modified autocorrelation method that is

Further, the voice synthesis method and voice synthesis apparatus according to each of the above-mentioned embodiments changes the pitch of the original voice to change the pitch of the original voice, that is, a pitch mark setting in a so-called voice pitch conversion apparatus. It is possible to adapt to processing in various audio output devices such as.

[0120]

As described in detail above, according to the present invention, the following effects can be obtained.

(1) Since the minimum point immediately before the peak of the audio signal is set as the center point of the audio signal to be cut out, the center point for superimposing can be easily set by a simple process and the spectral distortion is small. can do. As a result, the rumbling sound was reduced.

(2) By treating a voice segment with a fixed length as a unit and performing frame processing, it becomes easy to control voice waveform data at the time of voice synthesis.

(3) By preliminarily windowing the segments based on the obtained pitch period, there is no need to perform windowing processing during voice synthesis, and the amount of processing during voice synthesis processing can be greatly reduced. You can As a result, the processing device can be simplified or the processing speed can be increased.

(3) Since the function of inverting the positive / negative of the voice waveform is provided, it is possible to digitally correct the change in phase due to the change in the configuration of the analog system.

(4) Since the element is windowed in advance,
The processing amount at the time of speech synthesis processing can be greatly reduced.

[Brief description of drawings]

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

[Figure 2] PSOL that superimposes the voice waveform while changing the pitch
It is a schematic diagram which shows A method.

FIG. 3 is a flowchart showing a processing function in a segment creating unit of the speech synthesizer according to the first embodiment of the present invention.

FIG. 4 is a block diagram showing a configuration of a speech synthesizer according to a second embodiment of the present invention.

FIG. 5 is a flowchart showing a processing function in a segment creating unit of the speech synthesis device according to the second embodiment of the present invention.

FIG. 6 is a flowchart illustrating a cepstrum method.

FIG. 7 is a flowchart showing processing functions in a segment creating unit and an inversion control unit of a voice synthesizing device according to a third embodiment of the present invention.

FIG. 8 is a flowchart showing processing functions in a segment creating unit and an inversion control unit of a speech synthesis device according to a fourth embodiment of the present invention.

FIG. 9 is a block diagram showing a configuration of a voice synthesis device according to a fifth embodiment of the present invention.

FIG. 10 is a block diagram showing a configuration of a voice synthesis device according to a sixth embodiment of the present invention.

FIG. 11 is an explanatory diagram showing an example of detecting a minimum value immediately before a peak for voiced sound in each embodiment of the present invention.

[Explanation of symbols]

101: text analysis unit, 102: word dictionary, 103:
Parameter generation unit, 14: voice synthesis unit, 105: segment dictionary, 106: windowing unit, 107: segment creation unit, 108:
Audio signal input section.

Claims (10)

(57) [Claims] 1. A voice synthesis segment is created in advance by a step of detecting a local minimum point immediately before a peak of a voice signal, and a step of centering the detected local minimum point and cutting out the voice signal. A method for synthesizing speech, which comprises performing windowing superimposing while shifting a minimum point in a speech synthesizing unit as a center of superimposition and shifting by a pitch period. 2. A step of detecting a pitch period of an audio signal, a step of detecting a minimum point immediately before a peak of the audio signal, a step of centering the detected minimum point to cut out the audio signal, and a step of cutting out. A voice synthesis unit is created in advance by a step of multiplying the generated voice signal by a window that is a constant multiple of the pitch period, and the minimum point in the voice synthesis unit is used as the center of the superimposition to superimpose while shifting the pitch period. A voice synthesis method characterized by: 3. A step of appropriately inverting the positive and negative signs of a voice signal to match the positive and negative signs of the whole voice signal, a step of detecting a minimum point immediately before a peak of the voice signal, and centering around the detected minimum point. A voice synthesis characterized in that a voice synthesis unit is created in advance by the step of cutting out the voice signal, and the minimum point in the voice synthesis unit is used as a center of superposition to perform windowing superposition while shifting the pitch period. Method. 4. A step of detecting a pitch period of an audio signal, a step of appropriately inverting the positive and negative of the audio signal to match the positive and negative of the entire audio signal, and a step of detecting a local minimum point immediately before a peak of the audio signal. A voice synthesis unit is created in advance by a step of centering the detected minimum point and cutting out the voice signal, and a step of multiplying the cut out voice signal by a window that is a constant multiple of the pitch period. A method for synthesizing speech, which is characterized in that superimposing is performed while shifting a minimum point in a speech synthesizing unit as a center of superimposition and shifting by a pitch period. 5. A local minimum point detecting means for detecting a local minimum point immediately before a peak of an audio signal, and an audio signal clipping means for centering the local minimum point detected by the local minimum point detecting means to cut out the audio signal. A voice synthesis unit storage means for storing the voice synthesis unit cut out by the voice signal cutout unit, and a voice synthesis unit stored in the voice synthesis unit storage unit with the minimum point as the center of superimposition. A voice synthesizing apparatus comprising: a voice synthesizing unit that overlaps with a window while shifting the pitch period. 6. A pitch period detecting means for detecting a pitch period of an audio signal, a minimum point detecting means for detecting a minimum point immediately before a peak of the audio signal, and a minimum point detected by the minimum point detecting means. An audio signal cutting means for centering and cutting out the audio signal, a windowing means for applying a window having a constant multiple of the pitch period to the audio signal cut out by the audio signal cutting means, and a windowing means for applying the window A voice synthesis unit storing means for storing the voice synthesis unit, and a voice synthesis unit storing the voice synthesis unit stored in the voice synthesis unit storage unit while shifting the local minimum point as the center of superimposition while shifting the pitch period. A voice synthesizing device comprising: a synthesizing unit. 7. A voice signal inverting means for appropriately inverting the positive and negative signs of a voice signal to match the positive and negative signs of the whole voice signal, a minimum point detecting means for detecting a minimum point immediately before a peak of the voice signal, and the minimum point detecting means. A voice signal cutting-out means for cutting out the voice signal by centering on the minimum point detected by the above; and a voice synthesis unit storing means for storing the voice synthesis unit cut out by the voice signal cutting means, A voice synthesizing device comprising: a voice synthesizing unit for windowing and superimposing a voice synthesizing unit stored in a voice synthesizing unit storage means with a minimum point thereof as a center of superimposition and shifting by a pitch period. 8. A pitch period detecting means for detecting a pitch period of an audio signal, an audio signal inverting means for appropriately inverting the positive and negative of the audio signal to match the positive and negative of the entire audio signal, and a local minimum point immediately before a peak of the audio signal. Detecting means, a voice signal cutting-out means for centering the minimum point detected by the minimum-point detecting means to cut out the voice signal, and a voice signal cut-out by the voice signal cutting-out means Windowing means for applying a window that is a constant multiple of the pitch period, voice synthesis element storage means for storing the voice synthesis element windowed by the windowing means, and voice synthesis element storage means A voice synthesizing device, comprising: a voice synthesizing unit that superimposes a voice synthesizing element while shifting a minimum point of the voice synthesizing unit by a pitch period. 9. An inverting amplifier for inverting and amplifying a voice signal, a non-inverting amplifier for non-inverting amplification of a voice signal, a selector for selecting a voice signal from the inverting amplifier and a voice signal from the non-inverting amplifier, An AD converter that converts the audio signal selected by the selector into a digital value, a storage unit that stores the data AD-converted by the AD converter, and an audio signal that sequentially reads the audio signals stored in the storage unit. The reading means, the minimum point detecting means for detecting the minimum point immediately before the peak of the audio signal read by the audio signal reading means, and the minimum point detected by the minimum point detecting means are centered to center the audio signal. Audio signal cutout means for cutting out, and voice synthesis element storage means for storing the voice synthesis element cut out by the voice signal cutout means. A voice synthesizing apparatus comprising: a voice synthesis unit synthesizing unit for windowing and superimposing while shifting a pitch cycle about a minimum point of a voice synthesis unit selected from the voice synthesis unit storage means. . 10. An inverting amplifier for inverting and amplifying a voice signal, a non-inverting amplifier for non-inverting and amplifying a voice signal, a selector for selecting the voice signal from the inverting amplifier and the voice signal from the non-inverting amplifier, An AD converter that converts the audio signal selected by the selector into a digital value, a storage unit that stores the data AD-converted by the AD converter, and an audio signal that sequentially reads the audio signals stored in the storage unit. Reading means, pitch cycle detecting means for detecting the pitch cycle of the audio signal read by the audio signal reading means, window length calculating means for multiplying the pitch cycle detected by the pitch cycle detecting means by a constant, and reading the audio signal Minimum point detecting means for detecting the minimum point just before the peak of the audio signal read by the means, and the minimum point detected by the minimum point detecting means. An audio signal cutting-out means for centering the audio signal by centering on a point; windowing means for windowing the audio signal cut out by the audio signal cutting-out means with the window length calculated by the window length calculating section; A voice synthesis unit storage means for storing the voice synthesis unit windowed by the windowing unit, and a minimum point of the voice synthesis unit selected from the voice synthesis unit storage unit as a center of superposition, A voice synthesizing device, comprising: a segment connection synthesizing unit that overlaps while shifting by a pitch period.