JPH0727392B2 - Speech synthesizer - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION <Industrial field of application> The present invention relates to a speech synthesizer for generating a rule-synthesized speech.

<Prior Art> Making a predetermined vowel sound is important for generating a natural synthetic speech.

Conventionally, the rules for devoicing vowels are commonly used in “Shigeharu Sakurai“ Notes on Pronunciation of Common Languages ”Japanese Pronunciation Accent Dictionary (Revised New Edition), NHK, Commentary, Appendix P.128, 1985. It is shown as a law. It details a typical phonetic environment in which vowel devoicing occurs.

Further, as a rule of vowel devoicing which is actually used in a speech synthesizer, for example, “Yamato Sato, Kazuo Hakoda,“ Speech Synthesis by Law ”Research Practical Report, Vol. 27, No. 12, P. 2562.
(62) Denden Kosha 1978 ”.

FIG. 4 is a flowchart of a devoicing determination routine using the conventional vowel devoicing rule. The conventional unvoiced determination routine will be described below with reference to FIG.

In step S ₃₁ , the target vowel is a high tongue vowel (/ i /, / u /)
Is determined. As a result, if it is a high tongue vowel, the process proceeds to step S ₃₂ , and if not, it is determined to be voiced and the process proceeds to step S ₃₇ .

In step S ₃₂ , it is determined whether the target vowel is sandwiched by unvoiced consonants. As a result, the process proceeds to step S ₃₃ if it is pinched, step S it is determined that the voiced otherwise
Proceed to ₃₇ .

In step S _33, the vowel of interest accents nuclear whether it has the (pitch of a sound from a relatively high position changes to low) is determined. As a result, if it has an accent nucleus, it proceeds to step S ₃₉ , and if not, it proceeds to step S _34.
Proceed to.

In step S _34, whether or not a vowel of interest is the first mora is determined. As a result, if the first mora proceeds to step S _36, the process proceeds to step S ₃₅ otherwise.

In step S _35, whether or not the preceding vowel is already devoiced is determined. As a result, if unvoiced, it is regarded as semi-unvoiced, and the process proceeds to step S _39. If not, the process proceeds to step S ₃₆ .

In step S _36, whether or not a vowel of interest is sandwiched unvoiced fricative of the same type is determined. As a result, if it is sandwiched, it is regarded as semi-unvoiced and the process proceeds to step S ₃₉ , and if not, the process proceeds to step S ₃₈ .

In step S _37, it is determined to voicing the vowel of interest.

In step S _38, the vowels of interest is determined to be unvoiced.

In step S ₃₉ , quasi-unvoiced processing such as shortening the duration of the target vowel is executed.

<Problems to be Solved by the Invention> In actual speech, the vowel part may be completely unvoiced, but there are many phenomena in which a part of the vowel is devoiced and the rest remains voiced.

However, in the above-mentioned conventional unvoicing rule, although a quasi-unvoicing process for shortening the duration of a voiced vowel is performed on a vowel that is about to be unvoiced, it is essentially determined as a result of the unvoicing rule. Has only two states of unvoicing or voicing vowels, and no processing is performed to devoicize a part of a vowel and make the rest voiced. In addition, the rate of voiced unvoiced vowels generally changes when the utterance speed changes, but the conventional unvoicing rules quantitatively describe the degree of clear unvoicedness between the utterance speed and accent patterns and phonological sequences. Absent.

Therefore, the real voice and the unvoiced rule may not always match, and there is a problem that the synthesized voice generated based on the conventional unvoiced rule sounds unnatural.

Therefore, an object of the present invention is to obtain the degree of devoicing of vowels by phoneme, accent pattern, and vocalization rate based on the phoneme sequence, accent pattern, and vocalization rate. By calculating the unvoiced ratio of vowels and calculating the duration of voiced and unvoiced sections of vowels, it is possible to generate natural synthetic speech by changing the duration of unvoiced sections of vowels according to the actual speech. To provide a speech synthesizer.

<Means for Solving the Problems> In order to achieve the above-mentioned object, the present invention provides a speech synthesis parameter from the output of a character string analysis unit to which a character string is input according to a rule stored in a rule file in a speech synthesis parameter generation unit. In a voice synthesizer for performing voice synthesizing by a voice synthesizing means based on the voice synthesizing parameter, and a phonological sequence of a character string input from the character string analyzing unit and a phonological sequence stored in the rule file. Based on the degree of devoicing, the first devoicing degree setting means for setting the degree of dephonation of the phoneme sequence for each vowel of the input character string, and the character string input from the character string analysis unit. The accent pattern for each vowel of the input character string is based on the accent pattern and the degree of devoicing for each accent pattern stored in the rule file. The second devoicing degree setting means for setting the degree of devoicing and the degree of devoicing according to the designated speaking rate based on the degree of devoicing for each speaking rate stored in the rule file. A third devoicing degree setting means to be set, a phoneme sequence devoicing degree set by the first devoicing degree setting means, and an accent pattern devoicing degree set by the second devoicing degree setting means. ,
An unvoiced ratio setting means for obtaining an unvoiced ratio of vowels for each vowel based on the voicing speed unvoiced degree set by the third unvoiced degree setting means, and a speech synthesis parameter generated by the synthesis parameter generating means. Based on the duration of vowels, which is one of the vowels, and the unvoiced ratio of vowels obtained by the unvoiced ratio setting means, the duration and unvoiced interval of vowels of vowels for each vowel of the input character string. It is characterized by comprising a voiced / unvoiced duration length calculating means for calculating the duration of the.

<Operation> When an arbitrary character string is input to the character string analysis unit and the output from the character string analysis unit is input to the synthesis parameter generation unit,
The voice synthesis parameters are generated according to the rules stored in the rule file.

At that time, based on the phoneme sequence of the input character string and the degree of devoicing for each phoneme sequence stored in the rule file, the degree of dephonation of the phoneme sequence of each vowel of the input character string. Is set by the first unvoiced degree setting means. Also, based on the accent pattern of the input character string and the degree of devoicing for each accent pattern stored in the rule file, the accent pattern devoicing occurrence degree for each vowel of the input character string is second. It is set by the devoicing degree setting means. Further, based on the degree of devoicing for each speaking rate stored in the rule file, the degree of devoicing of the speaking rate is set by the third devoicing degree setting means according to the designated speaking rate.

Then, for each vowel, from the phoneme sequence devoicing degree set by the first, second and third devoicing degree setting means, the accent pattern devoicing degree, and the vocalization rate devoicing degree. The unvoiced ratio of vowels is obtained by the unvoiced ratio setting means. Then, based on the duration of vowels, which is one of the voice synthesis parameters generated by the synthesis parameter generation means, and the unvoiced ratio of vowels obtained by the unvoiced ratio setting means, the input character string For each vowel, the voiced and unvoiced duration of the vowel is calculated by the voiced / unvoiced duration calculating means.

After that, based on the voice synthesis parameter generated by the synthesis parameter generation means, the voiced section duration and the unvoiced section duration of the vowel calculated by the voiced / unvoiced duration length calculation means, Synthesis is performed.

Therefore, the voiced section duration and the unvoiced section duration of each vowel can be calculated in consideration of the phoneme sequence, the accent pattern, and the vocalization rate that have a great influence on the vowel devoicing phenomenon, and a natural synthesized speech can be obtained. it can.

<Embodiment> An outline of the configuration and operation of the speech synthesizer of the present invention will be described below with reference to the block diagram of FIG.

When an arbitrary character string is input to the character string analysis unit 1, the character string analysis unit 1 performs a syntactical interpretation of the input character string and determines the intonation pattern of the entire character string. Further, the word dictionary 2 is searched for a word included in the character string, and the accent and phonological sequence of each word in the character string is determined to determine the phonological sequence and accent pattern of the character string. In this way, the intonation pattern of the entire character string determined by the character string analysis unit 1 and the phoneme sequence and accent pattern of the character string are output to the rule control unit 3.

The characteristic parameter file 8 is composed of a target characteristic parameter file 6 and a time series characteristic parameter file 7, and the target characteristic parameter file 6 is
The target feature parameter representing the feature of the vowel is output to the rule control unit 3, and the time-series feature parameter file 7 outputs the time-series feature parameter representing the feature of the consonant to the rule control unit 3. On the other hand, the rule file 4 includes a phonological control rule for connecting the target characteristic parameter and the time-series characteristic parameter output from the characteristic parameter file, and a prosody control rule for controlling each prosody, respectively. Output to 3. The phonological control rules include phonological sequences, accent patterns, and devoicing coefficients that represent the degree of unvoicing for each utterance speed.

The rule control unit 3 controls the phoneme control rules and prosody for combining the target feature parameters and the time series feature parameters input from the feature parameter file 8 with the phonemes input from the rule file 4. With reference to the prosody control rule for performing the above, based on the intonation pattern of the entire character string, the phoneme sequence of the character string, and the accent pattern input from the character string analysis unit 1, the phonological feature parameters of the character string and the intervals between the phonemes. Of the pitch value, the power and duration of each phoneme, the voiced section duration and the unvoiced section duration of vowels, which will be described later, are generated, and the generated parameters are output to the speech synthesizer 5. To do.

The voice synthesizer 5 performs voice synthesis based on the input parameters and outputs a rule synthesizer corresponding to the input character string.

FIG. 2 is a flowchart of a vowel voiced / unvoiced section duration calculation routine executed by the rule control unit 3. The voiced segment / unvoiced duration calculation routine for vowels will be described below with reference to FIG.

In this routine, first, a devoicing coefficient indicating the degree of devoicing is calculated based on the phoneme sequence, accent pattern, and utterance speed output from the character string analysis unit 1.

In step S ₁ , the phoneme sequence devoicing coefficient ρp (n) based on the phoneme sequence is calculated as described in detail later.

However, n = 1, 2, ..., WM + 1 (WM is the number of mora for each phoneme symbol sequence input, and WM + 1 represents silence at the end). At step S ₂ , the accent pattern devoicing coefficient ρa
(N) is required.

The accent pattern devoicing coefficient ρa (n) is stored in the rule file 4 showing the devoicing coefficient based on various accent patterns depending on the accent type and the mora position of the target vowel, and is input according to this table. It is obtained by reading the devoicing coefficient of the accent pattern that matches the accent pattern of the character string. here,
The accent type is classified according to the position of a syllable that is highly sung among the syllables forming a word. For example, “n type” is a high syllable from the second syllable to the nth syllable. All syllables from the n + 1th onwards indicate that they are chanted low (Shinmei Kengoku Kanji Dictionary 3rd Edition Appendix Accent List).

The devoicing coefficient for each accent pattern has the following features.

1. Vowels in the accent nucleus are hardly devoiced.

2.0-type utterances have a higher degree of vowel devoicing than 1-type or more utterances.

3. Except for type 1 utterances, the beginning of words is highly unvoiced.

4. The vowel immediately before the accent kernel has a higher degree of devoicing than the vowel in the accent kernel, but has a lower degree of devoicing than the vowel after the accent kernel.

In step S ₃ , the vocalization speed devoicing coefficient ρs (n) is obtained.

As the vocalization rate devoicing coefficient ρs (n), a table showing devoicing coefficients based on various vocalization rates is stored in the rule file 4, and the devoicing coefficient of the specified vocalization rate is read from this table. Required by

In step S ₄ , the unvoiced ratio Rvu (n of the vowels is calculated by the following equation from the phoneme sequence devoicing coefficient, the accent pattern devoicing coefficient, and the vocalization rate devoicing coefficient obtained in steps S ₁ , S ₂ and S ₃ above. ) Is calculated.

Rvu (n) = ρp (n) × ρa (n) × ρs (n) However, when Rvu (n)> 1, Rvu (n) = 1 (1) where Rvu (n) = 1 (vowel The devoicing ratio of 100%) is the case of completely devoicing, and Rvu (n) = 0
The case where the unvoiced ratio of vowels is 0% is a case where no devoicing is performed. In other cases, for example, 0.75, the first 3/4 of the vowel following the consonant in the vowel section is unvoiced and the 1/4
Is the case of voicing.

In step S ₅ , the duration of the voiced section of the vowel and the duration of the unvoiced section are calculated as follows.

That is, in the duration of the vowel calculated by the rule control unit 3, the unvoiced ratio Rv of the vowel obtained in step S ₄ is calculated.
The unvoiced section duration of the vowel is calculated by multiplying u (n). After that, the duration of the vowel voiced section is calculated by subtracting the duration of the unvoiced section of the vowel from the duration of the vowel. However, when it is better to consider that the vowel is dropped, that is, when the duration of the vowel is 0, no matter what the ratio between the duration of the vowel and the unvoiced duration is, The duration of the unvoiced section is set to 0.

FIG. 3 is a flowchart of the above-mentioned phoneme sequence unvoiced coefficient ρp (n) calculation routine. Hereinafter, the phoneme sequence devoicing coefficient ρp (n) calculation routine will be described with reference to FIG.

In step S ₁₁ , a CVC (consonant-vowel-consonant) phoneme string centered on the target vowel is extracted from the phoneme sequence output from the character string analysis unit 1 for each vowel.

In step S _12, CVC devoicing coefficient for each CVC phoneme sequence is obtained.

For this CVC devoicing coefficient, a table showing devoicing coefficients based on the CVC phonological sequence is stored in the rule file 4, and the devoicing coefficient of the CVC phonological sequence that matches the CVC phonological sequence input from this table is read out. Required by.

In step S ₁₃ , n is used to make a determination from the first mora.
= 1.

In step S ₁₄ , the CVC de-voicing coefficient ρ cvc (n + 1) of the next mora (n + 1) and the CVC de-voicing coefficient ρ cv of the current mora (n).
It is determined whether or not at least one of c (n) is 0. As a result, if it is determined that one is 0, the process proceeds to step S ₁₈ , and if not, the process proceeds to step S ₁₅ .

In step S ₁₅ , the CVC de-voicing coefficient ρ cvc (n + 1) of the next mora (n + 1) and the CVC de-voicing coefficient ρ cv of the current mora (n).
c (n) is compared. As a result, ρcvc (n) ≧ ρcv
If c (n + 1), that is, the degree of devoicing of the next CVC phoneme sequence is low, it is determined that the degree of unvoicing of the next CVC phoneme sequence is low, and the process proceeds to step S ₁₇ . On the other hand, ρcvc (n)
<When ρcvc (n + 1) i.e. towards the degree of Devoicing is low the current CVC phoneme sequence proceeds to step S ₁₆ as the degree of devoicing current CVC phoneme sequence is reduced.

This is to make one vowel with a weak degree of devoicing devoiced in order to avoid unclear pronunciation of the vowels to be devoted subsequently.

In step S ₁₆ , the CVC devoicing coefficient of the current mora is calculated by the following equation.

ρcvc (n) = ρcvc (n) × β1 (β1 is a constant 0 ≦ β1 <1) That is, the CVC de-voicing coefficient of the current mora is reduced to make it difficult to devoice.

In step S ₁₇ , the CVC devoicing coefficient of the next mora is calculated by the following equation.

ρcvc (n + 1) = ρcvc (n + 1) × β2 (β2 is a constant of 0 ≦ β2 <1) That is, the CVC de-voicing coefficient of the next mora is reduced to make it difficult to devoice.

In step S ₁₈ , the phoneme sequence devoicing coefficient ρp (n) is calculated by the following equation.

ρp (n) = ρcvc (n) In step S ₁₉ , the phoneme sequence devoicing coefficient ρp of the next mora
In order to calculate (n + 1), n is incremented by 1.

In step S ₂₀ , n is the number of mora (W
M) It is determined whether or not the following. As a result, n ≦ WM
If so, steps S ₁₄ to S ₁₉ are repeated. on the other hand,
If n> WM, the phoneme sequence devoicing coefficient calculation routine ends.

In the above embodiment, when obtaining the unvoiced ratio Rvu (n) of vowels, an expression is calculated from the phoneme sequence devoicing coefficient ρp (n), the accent pattern devoicing coefficient ρa (n) and the vocalization rate devoicing coefficient ρs (n). It is calculated by (1). However, the three devoicing coefficients ρp (n), ρa (n),
A table showing the ratio of voiced sections or unvoiced sections of vowels based on ρs (n) may be stored in the rule file 4, and the unvoiced ratio Rvu (n) may be obtained from this table.

<Effects of the Invention> As is clear from the above, the speech synthesizer of the present invention is
Based on the degree of devoicing for each phoneme sequence, accent pattern, and speaking rate stored in the rule file,
The phoneme sequence, the accent pattern, and the voicing rate devoicing degree are set, the unvoiced ratio of vowels is obtained from the phonological sequence, the accent pattern, and the voicing rate devoicing degree, and the vowel duration and the vowel unvoiced ratio are used to determine the vowels. Since the duration of the voiced section and the duration of the unvoiced section are calculated, the duration of the unvoiced section of the vowel can be changed based on the phoneme sequence, the accent pattern, and the utterance speed. Therefore, it is possible to generate a natural synthesized voice according to the actual voice.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of a speech synthesizer of the present invention, and FIG. 2 is a vowel voiced section in the above embodiment.
Flowchart of unvoiced section duration calculation routine, third
FIG. 4 is a flowchart of the phoneme sequence devoicing coefficient calculation routine in the above embodiment, and FIG. 4 is a flowchart of the conventional devoicing determination routine. 1 ... Character string analysis unit, 2 ... Word dictionary, 3 ... Rule control unit, 4 ... Rule file, 5 ... Speech synthesizer, 6 ... Target feature parameter file, 7 ... Time-series feature parameter file , 8 ... Feature parameter file.

Claims (1)

[Claims] 1. A speech synthesis parameter is generated by a synthesis parameter generation means according to a rule stored in a rule file from an output of a character string analysis section to which a character string is input, and the speech synthesis means is based on this speech synthesis parameter. In a voice synthesizer for performing voice synthesis, a phoneme sequence of a character string input from the character string analysis unit,
First devoicing degree setting means for setting the degree of dephonization of the phoneme sequence for each vowel of the input character string based on the degree of devoicing for each phoneme sequence stored in the rule file; Based on the accent pattern of the character string input from the string analysis unit and the degree of devoicing for each accent pattern stored in the rule file, the degree of devoicing of the accent pattern of each vowel of the input character string. And a third devoicing degree setting means for setting the vocalization speed devoicing degree in accordance with the designated vocalization rate based on the vocalization rate-dependent devoicing degree stored in the rule file. Degree setting means, the degree of phoneme sequence devoicing set by the first unvoiced degree setting means, and the second unvoiced degree setting means An unvoiced ratio setting unit that obtains an unvoiced ratio of vowels for each vowel from the degree of unvoiced accent pattern and the degree of unvoiced speed set by the third unvoiced degree setting unit, and the synthesis parameter generation. The duration of vowels, which is one of the voice synthesis parameters generated by the means, and the unvoiced ratio of the vowels obtained by the unvoiced ratio setting means, based on the vowels of each vowel of the input character string. A voice synthesizer comprising a voiced / unvoiced duration length calculating means for calculating a duration of a voiced section and a duration of an unvoiced section.