JP3437472B2 - Speech synthesis method and apparatus - Google Patents

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 for high quality voice synthesizing, and more particularly to a voice synthesizing method and its device in which deterioration of the voice quality due to a fundamental frequency change is not noticeable.

[0002]

2. Description of the Related Art A conventional method for determining a fundamental frequency pattern in speech synthesis is to set the fundamental frequency of each syllable in consideration of the influence of the phonological environment, as described in JP-A-5-88690. There is known a method of connecting fundamental frequencies for each syllable to generate a fundamental frequency pattern of a syllable or a sentence.

[0003]

However, in the conventional method of determining the fundamental frequency pattern, the fundamental frequency pattern to be synthesized is determined without considering the fundamental frequency of the speech unit used for synthesis. There is a problem that the fundamental frequency change rate becomes large depending on the speech unit used, which causes deterioration of sound quality.

The present invention is intended to solve such a problem, and to reduce the fundamental frequency change rate of the speech unit within a range in which the intonation of the speech to be synthesized does not become unnatural. It is an object of the present invention to provide a speech synthesis method and apparatus capable of reducing deterioration of sound quality due to fundamental frequency change by determining a fundamental frequency pattern to be used.

[0005]

In order to solve the above problems, the present invention first prepares a typical fundamental frequency pattern capable of expressing the intonation of speech to be synthesized. Then, even if you change this basic frequency pattern,
Set a range that does not affect the intonation. Within the range in which the fundamental frequency pattern of the speech to be synthesized is variable, a fundamental frequency pattern for synthesis is determined so that the speech unit used for synthesis can be synthesized with a fundamental frequency change rate that is as small as possible. As a result, it is possible to obtain a voice synthesizing method capable of reducing the sound quality deterioration due to the fundamental frequency changing process while expressing the intonation of the voice to be synthesized.

[0006]

BEST MODE FOR CARRYING OUT THE INVENTION According to the invention described in claim 1 of the present invention, the fundamental frequency of a speech unit is set to
In speech synthesis method of the waveform superposition type for synthesizing speech into a fundamental frequency pattern, in determining the fundamental frequency pattern, to the extent that intonation does not become unnatural, the allowable range is provided to the fundamental frequency pattern, Previous
Serial to determine the fundamental frequency pattern in the I fundamental frequency is equal urchin before Symbol tolerance of each speech unit in the connection position of each speech unit, it is speech synthesis method characterized by synthesizing a speech By determining the fundamental frequency pattern to be synthesized so that the fundamental frequency change rate of the speech unit used for synthesis is reduced, it is possible to suppress the sound quality deterioration due to the fundamental frequency change.

The invention described in claim 2 is the same as claim 1
In the speech synthesis method described, the section for setting the allowable range of the basic frequency pattern is performed only for the steady section of the vowel part, and the basic frequency pattern in which the basic frequency change rate in the steady section of the vowel part is as small as possible is determined. This is a voice synthesis method characterized by that, for example, when a VCV unit is used as a unit used for synthesis, if a basic frequency pattern is determined only by a vowel part, a basic frequency pattern to be synthesized can be calculated easily. It has the effect that it can be set.

The invention described in claim 3 is the same as claim 1
The speech synthesis method described above is a speech synthesis method characterized in that an allowable range of a fundamental frequency pattern is determined for each type of phoneme to be synthesized, and a target fundamental frequency can be taken for each type of phoneme. By setting the allowable range, it is possible to reproduce the fundamental frequency pattern in consideration of the influence of the difference in phoneme.

The invention according to claim 4 is the same as claim 1
The speech synthesis method described above is a speech synthesis method characterized by determining an allowable range of a fundamental frequency pattern for each position of a syllable of a speech to be synthesized, and a fundamental frequency pattern such as a syllable in which a beginning or accent nucleus exists. By setting the permissible range of the fundamental frequency separately for syllables that have a large effect on the intonation and for syllables that do not, the deterioration of the sound quality due to the fundamental frequency change is less noticeable, and the fundamental frequency pattern with a natural intonation is created. It has the effect that it can be determined.

The invention according to claim 5 is the speech unit.
The fundamental frequency of the
Waveform superposition type speech synthesis that converts into turns and synthesizes speech
A device, which is a voice unit database for storing voice units in units of CV, VCV, CVC, etc., a character string input means for inputting a voice reading to be synthesized as a character string, and a voice unit using the character string. Within the range that does not affect the intonation of the voice that is the target of synthesis, a voice unit search means that searches for a candidate voice unit from the database ,
The allowable range is set in the basic frequency pattern,
At the connection position, the fundamental frequencies of the speech units will be equal.
As described above, a fundamental frequency pattern determining means for determining the fundamental frequency pattern within the allowable range, a fundamental frequency changing means for changing the fundamental frequency of each speech element according to the fundamental frequency pattern, and a speech element with the fundamental frequency changed A speech synthesis device comprising a speech unit connecting means to be connected and a synthetic speech output means for outputting a synthesized speech obtained by connecting the speech units, wherein the speech unit is stored in a speech unit database. Even if one side does not have a sufficient number of basic frequency patterns, it is possible to reduce the basic frequency change rate by freely setting the basic frequency pattern that is the synthesis target within the range where the intonation does not become unnatural. Therefore, there is an effect that it is possible to perform synthesis with little deterioration in sound quality without increasing the number of speech units in the speech unit database.

The embodiment of the present invention will be described below with reference to FIG.
It will be described with reference to FIG. (Embodiment 1) First, Embodiment 1 corresponding to the invention described in claim 1 of the present invention will be specifically described.
FIG. 1 is a conceptual diagram of the speech synthesis method according to claim 1.
In FIG. 1, 110 is a typical fundamental frequency pattern that expresses the intonation of speech to be synthesized, and 11
1, 112 are the upper and lower limits of the allowable range in which the basic frequency pattern can be set within the range where the intonation is not unnatural, 113 is the allowable range in which the basic frequency pattern can be changed within the range where the intonation is not unnatural, 12
1 to 124 are fundamental frequencies of speech units used for synthesis, 13
0 is a fundamental frequency pattern set so that the fundamental frequency change rate of each speech unit is small, and 141 to 144 are connection positions of each speech unit.

Next, the operation will be described with reference to FIG. In the synthesis of the waveform superimposing method, the fundamental frequency pattern 1 of each speech unit is added in order to give an appropriate intonation to the synthesized speech.
21 to 124 are changed to the fundamental frequency pattern 110 of the voice to be synthesized. As the representative fundamental frequency pattern 110, a fundamental frequency pattern obtained from actual speech is used, or a fundamental frequency pattern generated using a model based on the speech generation process is used. However, when changing the fundamental frequency, if the fundamental frequency patterns 121 to 124 of the respective speech units are completely changed to the fundamental frequency pattern 110 serving as the reference of the synthesized speech,
The fundamental frequency change rate becomes large, which may cause deterioration of sound quality. In the method of the present invention, the allowable range 113 in which the fundamental frequency pattern can be set is provided within the range where the intonation does not become unnatural. Within the basic frequency pattern setting allowable range 113, the basic frequency pattern 130 in which the basic frequency change rate of each speech unit is as small as possible is determined. At this time, the connection positions 141 to 14 of the respective speech units
In step 4, the basic frequency pattern is set so that it does not become discontinuous.

As described above, according to the first embodiment of the present invention, the basic frequency change rate of each speech unit is reduced within the allowable range in which the intonation does not make the basic frequency pattern of the synthetic target speech unnatural. By deciding
The effect that the sound quality deterioration due to the fundamental frequency change can be reduced can be obtained.

(Second Embodiment) Next, a second embodiment corresponding to the invention described in claim 2 of the present invention will be specifically described. FIG. 2 is a conceptual diagram of the speech synthesis method according to the second aspect. In FIG. 2, 201 is a basic frequency pattern of a synthesis target voice, 202 is a basic frequency pattern setting allowable range, 211 to 214 are basic frequency patterns of speech units used for synthesis, and 221 to 224 are basic vowel positions of speech units. It is the difference (fundamental frequency change rate) between the frequency and the fundamental frequency of the speech to be synthesized.

Next, the operation will be described with reference to FIG. Here, the operation will be described by using a voice unit whose connection unit is VCV (vowel-consonant-vowel). Similar to the first embodiment of the present invention, in the present invention, a target fundamental frequency pattern is set so that the fundamental frequency change rate is as small as possible within a range where the intonation does not become unnatural. However, in the case of using a speech unit with VCV as a connection unit for synthesis, if the consonant part is an unvoiced consonant, it has no fundamental frequency (211 and 214 in FIG. 2), and if it is a voiced consonant, the fundamental frequency. Is not stable (212, 21 in FIG. 2)
3) It is difficult to define the fundamental frequency change rate for the entire speech unit. Therefore, in the present invention according to claim 2, V
In the case of using a speech unit with a CV as a connection unit for synthesis, attention is paid only to the fundamental frequency change rate (221 to 224) in the steady section of the preceding vowel and the subsequent vowel, and the fundamental frequency change rate in this vowel steady section is small. The target fundamental frequency pattern is set so that Each vowel 1 as shown in FIG.
If attention is paid only to the points, the setting of the fundamental frequency pattern can be easily performed.

As described above, according to the second embodiment of the present invention, by focusing on the fundamental frequency change rate only at each one point of the preceding vowel and the following vowel in the VCV segment,
The effect that the target fundamental frequency pattern can be easily set is obtained. In the above description, the case where VCV is used as a connection unit and the fundamental frequency change rate is determined with one vowel to synthesize the word voice is explained. However, when synthesizing a phrase or a sentence, a natural fundamental frequency pattern is similarly obtained. Can be set.

(Third Embodiment) Next, a third embodiment corresponding to the invention described in claim 3 of the present invention will be specifically described. FIG. 3 is a conceptual diagram of the speech synthesis method according to the third aspect. In FIG. 3, 310 and 320 are typical fundamental frequency patterns of speech to be synthesized, 311 to 31 1.
3, 321 to 323 are ranges in which setting of the fundamental frequency is permitted.

Next, the operation will be described with reference to FIG. When synthesizing the word voices "change" and "culture" shown in FIG. 3, since the two words have the same accent type of 3 moras,
Representative fundamental frequency patterns 310 and 320 having the same shape
Are used respectively. However, even with words having the same accent type, the fundamental frequency pattern may differ due to the difference in phoneme. (Generally, the fundamental frequency rises from a low frequency in voiced consonants, whereas it tends to start from a high frequency in unvoiced consonants.) Therefore, the word "change" in which the first syllable is the unvoiced consonant "he". , And the word “culture” where the first syllable is the voiced consonant “bu”, in consideration of the influence of each consonant, the allowable range 311 is set higher in the case of “he” and in the case of “bu”. Is a lower tolerance 321
To set.

As described above, according to the third embodiment of the present invention, considering the influence of the difference in phoneme on the fundamental frequency,
By providing the correction allowable range of the fundamental frequency pattern for each phoneme, the fundamental frequency pattern closer to the original voice can be reproduced, and the effect of reducing the fundamental frequency change rate can be obtained. In the above description, the case where the word voice is used has been described, but a more natural fundamental frequency pattern can be set by considering the influence of the phoneme even in the phrase and the sentence.

(Embodiment 4) Next, Embodiment 4 corresponding to the invention described in claim 4 of the present invention will be specifically described. FIG. 4 is a conceptual diagram of the speech synthesis method according to claim 4. In FIG. 4, reference numeral 401 is a typical basic frequency pattern of a voice to be synthesized, and 411 to 416 are ranges in which the setting of the basic frequency pattern set for each mora is allowed.

Next, the operation will be described with reference to FIG. FIG. 4 is an example of a fundamental frequency pattern used when synthesizing a 6-mora type 3 accent word. In the fundamental frequency pattern 401 of the speech to be synthesized, the second syllable in which the fundamental frequency reaches the maximum value and the third syllable in which the accent nucleus exists are large factors that determine the accent type of the word speech, so Detailed settings are required. Therefore, in the method of the present invention, not all syllables give the same setting tolerance of the fundamental frequency, but the setting tolerance of the fundamental frequency is set for each syllable. That is, since the influence on the intonation is strong near the accent nucleus and the target fundamental frequency pattern is important, the setting allowable range 4
12, 413 is set to be narrower, and in syllables after the accent nucleus, it is not necessary to set the fundamental frequency pattern as strictly as in the accent nucleus, so that the allowable range 414 to 41 is set so as to reduce the sound quality deterioration due to the fundamental frequency change.
Widely set 6

As described above, according to the fourth embodiment of the present invention, the basic frequency setting permissible range is separately provided for each syllable, so that the basic frequency can be reduced while the basic frequency change rate is reduced. The effect that the pattern can be determined is obtained. In the above description, the case where the word voice is used has been described, but it is possible to set a more natural fundamental frequency pattern by considering the position of the syllable in a phrase or a sentence.

(Fifth Embodiment) Next, a fifth embodiment corresponding to the invention described in claim 5 of the present invention will be specifically described. FIG. 5 is a block diagram of a speech synthesizer according to claim 5. In FIG. 5, 501 is a character string representing the reading of the speech to be synthesized, 502 is a character string input means for inputting this character string, and 503 is a candidate speech unit from the speech unit database according to the input character string. A speech element search means for searching, 504 is a fundamental frequency pattern determining means for determining a fundamental frequency pattern and a speech element to be synthesized, and 505 is a fundamental frequency changing means for changing the fundamental frequency of each speech element according to the fundamental frequency pattern. 506 is a voice unit connecting means for connecting voice units whose fundamental frequency has been changed, 507 is a synthesized voice output unit for outputting a synthesized voice obtained by connecting the voice units, 508 is an output synthesized voice, A voice unit database 509 stores voice units in units such as CV, VCV, and CVC.

Next, the operation will be described with reference to FIG. First, a character string containing information such as accent marks in the reading of the synthetic target voice is input from the character string input means 502 and decomposed into CV or VCV units. Next, a plurality of speech units having the same reading as the decomposed character string are searched from the speech unit database 509 by the speech unit searching unit 503. From among the searched candidate speech units, a speech unit whose fundamental frequency change rate of the speech unit is as small as possible is determined by comparing with the fundamental frequency changeable range obtained by the fundamental frequency pattern determination unit 504. . Also, at the same time as determining the speech unit, within a range where the intonation does not become unnatural,
A fundamental frequency pattern in which the fundamental frequency change rate is as small as possible is determined based on the methods according to claims 1 to 4 of the present invention. According to the fundamental frequency pattern determined here,
Basic frequency changing means 50 is used to change the basic frequency of each speech unit.
Do in 5. Finally, the voice units whose fundamental frequencies have been changed are connected by the voice unit connecting means 506 to output a synthesized voice.

As described above, according to the fifth embodiment of the present invention, even if a sufficient number of types of basic frequency patterns are not prepared for the speech units stored in the speech unit database,
Since the fundamental frequency change rate can be reduced by modifying the fundamental frequency pattern that is the synthesis target within the range where the intonation does not become unnatural, there is little deterioration in sound quality without increasing the number of speech units in the speech unit database. The effect that a voice synthesizer can be configured is obtained.

[0026]

As is apparent from the above embodiment, the present invention focuses on the fundamental frequency of the speech unit used for synthesis,
By deciding the fundamental frequency pattern to be synthesized so that the fundamental frequency change rate of each speech unit is reduced within a range where the intonation does not become unnatural, a speech synthesis method and its device that can reduce sound quality deterioration are provided. Obtainable.

[Brief description of drawings]

FIG. 1 is a conceptual diagram showing an operation of a speech synthesis method according to a first embodiment of the present invention.

FIG. 2 is a conceptual diagram showing an operation of a fundamental frequency pattern determination method focusing only on a vowel position in the second embodiment of the present invention.

FIG. 3 is a conceptual diagram showing a method of determining a permissible range in which a fundamental frequency can be set according to a third embodiment of the present invention.

FIG. 4 is a conceptual diagram showing a method for determining a permissible range in which a fundamental frequency can be set according to Embodiment 4 of the present invention.

FIG. 5 is a block diagram showing a configuration of a speech synthesizer according to a fifth embodiment of the present invention.

[Explanation of symbols]

110, 310, 320, 401 Typical basic frequency patterns 121 to 124, 211 to 214 Basic frequencies 130 and 201 of speech units Basic frequency patterns 141 to 144 for synthesis purpose Connection position of each speech unit 202 Basic frequency pattern setting Allowable range 502 Character string input unit 503 Speech unit search unit 504 Basic frequency pattern determination unit 505 Basic frequency change unit 506 Speech unit connection unit 507 Synthesized speech output unit 509 Speech unit database

─────────────────────────────────────────────────── --- Continuation of the front page (56) References JP-A-1-284898 (JP, A) JP-A-4-281499 (JP, A) JP-A-8-254993 (JP, A) JP-A-10- 97291 (JP, A) (58) Fields surveyed (Int.Cl. ⁷ , DB name) G10L 13/08

Claims (5)

(57) [Claims] 1. A fundamental frequency of a speech unit is used for the purpose of synthesis.
In the waveform superposition type speech synthesis method for synthesizing a speech is converted to the fundamental frequency pattern of voice that, in determining the fundamental frequency pattern, to the extent that intonation is not unnatural, tolerance to the fundamental frequency pattern the provided, wherein the connecting position of each speech unit each speech segment
Fundamental frequency determines the <br/> fundamental frequency pattern in the I urchin before Symbol tolerance equal, speech synthesis method characterized by synthesizing the speech. 2. The speech synthesis method according to claim 1, wherein
Speech synthesis characterized by performing a section for setting an allowable range of the basic frequency pattern only for a stationary section of a vowel part, and determining a basic frequency pattern in which a basic frequency change rate in the stationary section of the vowel part is as small as possible. Method. 3. The speech synthesis method according to claim 1, wherein
A method for synthesizing speech, characterized in that an allowable range of a fundamental frequency pattern is determined for each type of phoneme of speech to be synthesized. 4. The speech synthesis method according to claim 1, wherein
A method for synthesizing a voice, characterized in that an allowable range of a basic frequency pattern is determined for each position of a syllable of a voice to be synthesized. 5. The fundamental frequency of a speech unit is used for the purpose of synthesis.
Convert to the fundamental frequency pattern of the voice to synthesize the voice
A waveform superimposing type speech synthesizer, which is a CV, VCV,
A speech unit database that stores speech units in units such as CVC, character string input means that inputs the reading of synthesized speech as a character string, and speech units that are candidates from the speech unit database by the character string. An allowable range for the fundamental frequency pattern within the range that does not affect the intonation of the speech unit to be searched and the speech to be synthesized.
Is provided, and at the connection position of the speech unit, the base of the speech unit is
Basic frequency pattern determining means for determining the basic frequency pattern within the permissible range so that the main frequencies are equal, basic frequency changing means for changing the basic frequency of each speech element according to the basic frequency pattern, and the basic frequency A voice synthesizing apparatus comprising: a voice unit connecting means for connecting a changed voice unit; and a synthesized voice output unit for outputting a synthesized voice obtained by connecting the voice units.