JPH0990972A - Synthesis unit generating method for voice synthesis - Google Patents

Abstract

PROBLEM TO BE SOLVED: To obtain smooth and stable synthesized voices by commonly using the phoneme environment, which is equivalent in terms of phonetics, in the voice segment generated while considering the effect of a sound adjustment style based on the phoneme environment and generating a synthesis unit. SOLUTION: In a step S1, phoneme environment such as a sound adjustment combination is commonly used if it is considered to be same (being considered to be the same) so as to reduce the number of synthesis units, while considering phoneme origins. In a step S2, if the succeeding phoneme of the phoneme (immediately after the phoneme) is a long vowel, the phoneme environment of the phoneme is considered to be the phoneme environment in which a short vowel is the succeeding phoneme. In other word, a long vowel is considered to be same as a short vowel in the phoneme environment of the succeeding phoneme. In a step S3, if the phoneme is a voiceless bursting sound, the preceeding phoneme is represented by one short vowel. In a step S4, if the succeeding phoneme of the phoneme is a voiceless bursting sound, the succeeding phoneme 15 represented by one voiceless plosive. Thus, by all of the steps S1 through S4, a group is made and the number of phoneme is reduced.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for creating a speech synthesis unit in speech synthesis for artificially generating speech.

[0002]

2. Description of the Related Art Conventionally, in a regular voice synthesis system, when "C" is a consonant and "V" is a vowel, VCV, CV, CVC, etc. are often used as a synthesis unit. Compared to the method using such a composition unit,
The rule-based speech synthesis method, which uses the phoneme, which is the most basic unit of speech, as the synthesis unit, makes the connection rules at the time of synthesis complicated even though it is natural to use phonemes as the synthesis unit. It was not used.

Conventionally, when a phoneme is used as a synthesis unit for regular speech synthesis, the number of units becomes enormous. Therefore, the preceding and following phonological environments are not considered at all, or the COC (context-
oriented-clustering) method (see Japanese Patent Application No. 62-23 for details)
The application of No. 4358 is to be applied either (see attached description and drawings).

[0004]

By the way, when a phoneme that does not consider the preceding and following phonological environments is used as the synthesis unit for the regular speech synthesis, the number of synthesis units is small and the difference in expression depending on the language is small. Although it has the advantage that it can be used in a language, it is difficult to obtain high-quality synthesized speech because the transition part from the consonant (C) to the vowel (V), which has a particularly large change in the speech characteristics, is also connected by a rule. there were.

On the other hand, the COC method is a method of clustering phonemes considering a phonological environment (articulatory coupling) from a large population of voice data by a statistical method and accumulating the centroid as a synthesis unit. The COC method has an advantage that a synthetic unit is generated based on an objective scale because clustering is automatically performed. However, it takes a lot of man-hours to maintain a clustering population, and a text for creating a synthetic unit ( There is a possibility that there is a phonological environment that does not appear in the population because the (voice data) is finite.

Further, since a large amount of synthesis unit tables are searched for the optimum synthesis unit at the time of speech synthesis, there is a disadvantage that the table search takes time. Furthermore, since a synthesis unit is generated from natural utterances and accumulated to form a population, there are large variations due to the frequency of occurrence of centroids, and during speech synthesis it is optimal even if the fitness to the phoneme environment is small. A candidate may be selected as a new synthesis unit. Therefore, there is a drawback that the quality of synthesized speech is unstable. In order to solve these drawbacks, it is possible to consider a method in which a phoneme is taken into consideration in consideration of all the phonetic environments before and after, but it is not realistic because the number of synthesis units becomes huge.

As described above, in regular speech synthesis, when the synthesis unit is a phoneme, the number of units can be reduced by using a phoneme that does not consider the preceding and following phoneme environments as the synthesis unit, but a smooth synthesized speech can be obtained. There wasn't. Further, when the COC method is used, it is not possible to expect stable generation of synthetic speech at all times. Furthermore, if a phoneme that takes into consideration all of the phonemic environments before and after is used as a synthesis unit, smoothness and stability may be compensated, but the number of synthesis units becomes huge, which is not realistic.

The present invention has been made in view of the above circumstances, and a first object of the present invention is to create a synthesis unit that substantially uniformly covers all phonemes in consideration of preceding and following phoneme environments,
It is an object of the present invention to provide a method for creating a synthesis unit for speech synthesis that can obtain smooth and stable synthesized speech. In addition, a second object of the present invention is to provide a synthesis unit creation method for speech synthesis which can reduce the number of synthesis units without deteriorating the quality of the obtained synthesized speech.

[0009]

According to the method for creating a synthesis unit for speech synthesis according to the present invention, in the method for creating a synthesis unit for speech synthesis, which creates a speech segment including a phoneme as a synthesis unit, the influence of the articulation style based on the phonological environment. It is characterized in that a voice segment is created in consideration of the above, and a phoneme environment that is phonetically equivalent is shared in the created voice segment to create a synthesis unit.

[0010]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Before describing the embodiments of the present invention, the basic concept of the present invention will be described. The present invention considers all speech segments (actually, phonemes (triphones)) in consideration of phonemic environments before and after appearing in a language such as Japanese, and selects the phoneme environment from among them in consideration of phoneme characteristics. It aims to reduce the number of units by grouping visible ones. Therefore, it is obviously different from the conventional technique in which the phonological environment is not considered or the synthesis unit is generated from the population by a statistical method.

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a diagram for explaining a method for creating a speech synthesis unit according to an embodiment of the present invention.
As shown in this figure, the present embodiment includes steps S1 to S1.
This is done by sequentially executing S4. The method for creating a synthesis unit for speech synthesis shown in FIG. 1 is used for Japanese speech synthesis, and each step will be described below in order. However, in the following description, c,
c ', C, C', Q, v, v ', V, V', V "respectively represent phonemes. Specifically, v, v'are long vowels, C, C ', c, c'is a consonant, C'is an unvoiced consonant of C, c, c'is an unvoiced explosive consonant, Q is a consonant, V, V ', V "are vowels, and V, V'are These represent v and v'short vowels, respectively.

[Step S1] In step S1, when the preceding phoneme of the phoneme of interest is a long vowel, the phoneme environment of the phoneme is a phoneme environment in which a short vowel is a preceding phoneme (previous phoneme). I reckon. That is, long vowels are identified with short vowels in the phoneme environment of the preceding phoneme. If all the triphones existing in Japanese are used as synthesis units, the number of units including consonants is about 15,000, which is an enormous number. Therefore, in step S1, in consideration of phoneme features, common identifiable phoneme environments such as articulatory combinations are identified (identified) to reduce the number of synthesis units. Below, the example of the same identification performed in step S1 is shown. Example 1: C of vCV 'is considered to be the same synthetic unit as C of VCV'. Example 2: Consider the C of vCC 'as the same synthetic unit as the C of VCC'. Example 3: Consider the v of v'vC as the same synthetic unit as the v of V'vC. The rules of Examples 1-3 above reduce about 3000 triphones.

[Step S2] In step S2, when the subsequent phoneme (immediate phoneme) of the phoneme is a long vowel, the phoneme environment of the phoneme is regarded as a phoneme environment in which a short vowel is a subsequent phoneme. That is, long vowels are identified with short vowels in the phoneme environment of the subsequent phoneme. Below, step S2
An example of the same identification performed in FIG. Example 4: Consider the C of V'Cv as the same synthetic unit as the C of V'CV. Example 5: V of CVv 'is considered to be the same synthetic unit as V of CVV'. Example 6: The v of V "vv 'is considered to be the same synthetic unit as the v of V"vV'. The rules of Examples 4-6 above reduce about 2000 triphones.

[Step S3] In step S3, when the phoneme is an unvoiced plosive, the preceding phoneme is represented by one short vowel. Below, an example of the identification performed in step S3 is shown. Example 7: c of VcV ″ is considered to be the same synthetic unit as c of V′CV ″. Example 8: C in VcC (unvoiced c) is considered to be the same synthesis unit as c in V'cC. Example 9: c of QcV is regarded as the same synthetic unit as c of V'cV. In Examples 7 to 9, V'represents a representative short vowel. Rules such as Examples 7-9 above reduce about 1000 triphones.

[Step S4] In step S4, when the succeeding phoneme of the phoneme is an unvoiced plosive sound, the succeeding phoneme is represented by one unvoiced plosive sound. Example 10: V of CVc is considered to be the same synthetic unit as V of CVc '. Example 11 FIG. C of VCc (unvoiced C) is regarded as the same synthesis unit as C of VCc '. In Examples 10 and 11, c'represents a representative unvoiced explosive consonant. Rules such as Examples 10 and 11 above reduce about 3000 triphones.

After all, when grouping is performed by all of the above-mentioned steps S1 to S4, about 9000 triphones are reduced, and about 600 triphones are combined.
It becomes 0 species. Although the embodiments of the present invention have been described in detail above with reference to the drawings, the specific identification rules are not limited to those exemplified in the above-described embodiments, and do not depart from the gist of the present invention. Even if there is a change in the range, it is included in the present invention.

[0017]

As described above, according to the present invention,
Since all triphones appearing in a predetermined language are taken into consideration, uniform and stable synthesis units can be accumulated and configured regardless of the frequency of appearance. In addition, the number of synthesis units can be significantly reduced by grouping adjacent (front and back) phoneme environments. For example, in Japanese, the number of synthesis units can be greatly reduced to 2/5 depending on the phoneme characteristics, and moreover, all triphones existing in Japanese can be handled substantially uniformly. Also, speech segments (phonemes) that take into consideration adjacent (preceding and following) phonological environments.
Is used as a synthesizing unit, it is possible to smoothly connect the voices when synthesizing the voices, and it is possible to synthesize a voice having excellent naturalness.

[Brief description of drawings]

FIG. 1 is a diagram illustrating a method of creating a synthesis unit for speech synthesis according to an embodiment of the present invention.

 ─────────────────────────────────────────────────── ─── Continued Front Page (72) Inventor Kenzo Ito 1-1-6 Uchisaiwaicho, Chiyoda-ku, Tokyo Nihon Telegraph and Telephone Corporation

Claims (1)

[Claims] 1. A method for creating a synthesis unit for speech synthesis, which creates a speech segment containing a phoneme as a synthesis unit, creates a speech segment in consideration of the influence of an articulation style based on a phoneme environment, and creates a speech segment in the created speech segment. A method for creating a synthesis unit for speech synthesis, which comprises creating a synthesis unit by commonizing phonetically equivalent phonological environments.