JPH01120600A - Voice rule synthesization - Google Patents

Abstract

PURPOSE: To prevent the degradation in tone quality of nasal sounds to obtain a synthesized voice of high comprehensibility by connecting a CV(consonant- vowel chain) type synthesis fundamental unit by a specific method at the time of discriminating a specific nosal sound. CONSTITUTION: A syllabic code is inputted to discriminate whether a consonant C is a specific nosal sound like m, n, my, or ny or not. If it is discriminated that the consonant is a specific nosal sound, the value of the CV type synthesis fundamental unit is cut into the duration of the nasal sound if this duration is longer than a preliminarily determined set duration, and the value representing the unit is inserted to the front space so that the duration of the nosal sound is equalized to the set duration if it is shorter than the set duration. The value of the vowel part just preceding the nosal sound is cut into the set duration if the duration of this vowel part is longer than the preliminarily determined set duration, and the value representing the vowel part of the unit is inserted to the rear space so that the duration is equalized to the set duration if the duration of the vowel part is shorter than the set duration. Thus, the degradation in tone quality of nosal sounds is prevented.

Description

DETAILED DESCRIPTION OF THE INVENTION [Industrial Field of Application] The present invention relates to a method of rule-based synthesis of text speech, and particularly to improving the sound quality of rule-synthesized speech.

[Conventional technology]

A method of synthesizing speech corresponding to an arbitrary sentence or word is called "speech synthesis by rules" or simply "speech synthesis by rules."

Speech synthesized by rules generally differs from natural speech because features such as phonological connections, duration, and changes in pitch (voice height) are externally imparted by rules. Therefore, the speech generated by rule synthesis is
One of the causes of the deterioration of the sound quality of rule-synthesized speech, which is worse than the sound quality of so-called "analytical synthesis" speech that preserves the characteristics of natural speech, is due to a decrease in the clarity of phonemes. Causes of decreased phonological intelligibility include:
Possible causes include the method of creating (analysis) the composite unit and the unit persistence process.

Among these, the present invention aims to improve the deterioration in sound quality caused by unit connection processing.

Traditionally, linear interpolation of parameters was mainly used to connect Cv units, as described in ``Speech synthesis using rev-1 clauses as units'' and the Proceedings of the Acoustical Society of Japan Conference Proceedings 3-4-3 (1982). . This linear interpolation was considered to be an important process for smoothing the changes in parameters between syllables, resulting in a synthesized speech with smooth connections.

[Problem that the invention seeks to solve]

The above is certainly true for many consonants, but in the case of nasal consonants such as /m/, /n/, /my/, /ny/, smoothing reduces the characteristics of these nasals. A certain "crispness" is lost, and intelligibility decreases.

An object of the present invention is to provide a unit conjunctive method that prevents such deterioration in the sound quality of nasal sounds, and to obtain synthesized speech with high intelligibility.

[Means for solving problems]

A unit connection method effective for solving the above problems will be explained below. What parameters are used to constitute the unit differs depending on the synthesis method.

For example, when the PARCOR synthesis method is used, the parameters are a predetermined order PARCOR coefficient and the sound source amplitude, and when the waveform synthesis method is used, the waveform amplitude value itself is the parameter. The present invention provides a general method that can be applied to any synthesis method, but here,
In order to make the explanation concrete, a synthesis method using residual compression method (Japanese Patent Application No. 58-42169) will be taken as an example. That is, FAI? as a spectral envelope parameter? This is a synthesis method that uses representative residuals (residuals for one pitch in the case of voiced, residuals for one frame in the case of unvoiced) as COR coefficients and sound source parameters.

FIG. 1 shows the principle of the unit connection method of the present invention. In FIG. 1, (a) shows the flow of the unit connection process.

That is, based on the input syllable information, it is determined whether the target syllable is "ma", "na", "miya", or "niya".
Determine the stage. If this syllable is one of these, apply the nasal connection process explained in (b) of the same figure, and
! If neither is the case, conventional linear interpolation processing is applied.

The nasal connection process will be explained below. Figure 1(b) shows the connection principle of a certain parameter, and this parameter is the first-order (i = 1.2°...p:p is the order of linear prediction) PARCOR coefficient, the representative One of the residuals. However, in the former case, processing is performed for each connected frame, and in the latter case, processing is performed for each corresponding sample point within the frame. Also, in this figure, -1 for the vowel V in the preceding C■ unit is shorter than the set duration length, and the corresponding Cv
An example is taken in which the consonant (nasal sound) CK in the unit is longer than the set duration length. In this case, the portion longer than the set time length of Cyc is truncated. Also, in the gap portion shorter than the set time length of V-1, the parameter value of the last frame of V-1 is filled. By combining these cutting and embedding processes so that there is no gap between C'', -t V, -1 and CKV, the nasal connection process is completed. Note that c v I', Even if the combination of the length of the digit j and the set duration length is different from this example, if the unit is longer, the unit is cut, and if it is shorter, the gap is used as the final value of the unit (
By performing embedding with the first value in the case of a succeeding unit and the last value in the case of a preceding unit, desired nasal connection processing can be performed.

In the above example, the amplitude is also discontinuous. Therefore, if you want to avoid discontinuity in the listening sensation, you can linearly interpolate only the sound source information (representative residual in this example).If linear interpolation is too smooth, then after the nasal sound processing described above, Amplitude weighting may be performed using a curve having an appropriate envelope shape. Such weighting can be realized by multiplying the time function of the weight of the predetermined envelope shape by the sound source waveform amplitude value.

[Effect]

Using the above unit conjunction, /m8 to n/, /■y/, /
For nasal sounds such as ny/, you can use conventional linear interpolation without losing their characteristics, and for other consonants, you can use traditional linear interpolation, so you can use the natural characteristics unique to each phoneme without losing the smoothness of the connections. It is possible to obtain synthetic speech.

〔Example〕

An embodiment of the present invention will be described below with reference to FIG.

Figure 2 shows the overall configuration of the arbitrary text synthesis method.

In this method, if a text containing kanji and kana is given as input data, the corresponding synthesized speech can be obtained as output. The processing procedure is as follows.

First, the input text is broken down into each word by the morphological analysis means of the Japanese language analysis unit 1, one part of speech is determined, and the pronunciation is determined. 59-13040, patent application 1987-1
90861. JP-A-59-126841), the accent type of each word or phrase is determined. Syllable information, accent information, etc. are obtained as a result of the above syntactic level processing. Note that the breaks between phrases and sentences are determined based on delimiters such as punctuation marks in the input text.0 The length of pauses in sentences and between sentences can be specified by the number of spaces after commas or periods. Also interrogative and imperative sentences.

The type of sentence, such as a desire sentence, can sometimes be determined by the conjugation of the ending of the word, or by using a final symbol such as "?", "!!", and "!" at the end of the sentence instead of a period, respectively. It can also be specified by using 0. For example, even if they have the same phonetic sequence ``Kawa wo Cross.'', ``Kawa wo Cross.'' is a declarative sentence, and r? ” is an interrogative sentence. ■Syllable information, ■Accent information, ■Pause information, ■Phrase/
Sentence break information and (if necessary, part-of-speech basis, etc.) ■grammatical information are expressed by a series of numbers called a "syllable code." The syllable code is input information to the control parameter generation section 3.

In the control parameter generation unit 3, accent, intonation, and phoneme duration are determined according to rules, and
Pitch bangs and phonological parameter time series are generated accordingly. Here, the accent type can be known from accent information. Specifically, the accent information is given by inserting a syllable code number indicating the accent immediately after the phoneme with the accent core (the sound fiA immediately before the accent falls). However, the absence of this syllable code indicates a flat accent. Furthermore, intonation is basically determined by sentence type information. However, variations may be added due to differences in the phonological order of the first word.For example, the intonation and slam are different between the wishful sentences ``I want to cross the river!'' and ``I want to cross the river!''.

The final pitch pattern is generated based on both accent type and intonation. In the case of consonants, the phoneme duration is determined as a unique length for each type of consonant since it is less affected by surrounding conditions.

In contrast, vowels undergo various transformations depending on surrounding conditions. Therefore, the duration is determined based on the accent type, the number of syllables, the position within the word, the type of consonant immediately before it, the type of vowel, etc. (see patent application 190861/1986). In this way, the phonological duration is Once determined, Cv
Phonological parameters (spectral envelope parameters and sound source parameters) registered in a file in units are extracted and arranged in correspondence with syllable codes. At this time, if it is too long, it is cut to fit within the duration. Thereafter, the Cv units are connected so as to fill the cut portion or gap.

The subjunctive method is a [means for solving problems] in this specification.
Use the method described in . Finally, the fundamental frequency and phoneme parameters generated through the above processing are sequentially sent to the speech synthesis section 4, and a speech waveform is output. Here, as a speech synthesis method, one example is the residual compression method (Japanese patent application No. 59-5
583118, Japanese Patent Application No. 60-137721) may be used. in this case.

Sound source pulses are basically generated by extracting residual pulses (representative residuals) of one pitch sentence for each frame and arranging the representative residuals at intervals of a pitch period given from the outside. At this time, if the pitch period given from the outside is shorter than the length of the representative residual, the end of the representative residual is truncated by that length difference, and if it is longer, 0 is filled in for the section where the representative residual is insufficient. There is. The amplitude values of the residual pulse trains arranged in this manner are subjected to appropriate gland-shaped weighting or the like in order to smooth the connections between frames.

All of the above processing can be configured by known means, except for the Cv unit connection method, which is the core of the present invention.

According to this embodiment, it is possible to obtain synthesized speech with high intelligibility from a text containing kanji and kana.

In particular, it has the effect of improving the intelligibility of nasal sounds.

〔Effect of the invention〕

As shown above, the present invention is effective in improving the clarity of nasal sounds and is effective in improving the sound quality of rule-synthesized speech.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a diagram showing the basic part of the present invention, and FIG. 2 is a diagram showing an embodiment of the present invention. 3... Control parameter generation section.
r7=,,. Kaya 1 Figure (Hisashi) (Reno

Claims (1)

[Claims] 1. As a means for connecting CV (consonant-vowel chain) type synthesis basic units (hereinafter referred to as "units"), C is /
It is determined whether the nasal sound is a specific nasal sound such as m/, /n/, /my/, /ny/, etc., and when the determining means determines that the nasal sound is a specific nasal sound, the duration of the nasal sound is determined in advance. If it is longer than the set duration length, the value of the above unit is cut to the set duration length, and conversely, if the duration length of the nasal sound is shorter than the set duration length, a value representative of the above unit is cut to the set duration length. It is filled in the gap in front so as to match the set duration length, and if the duration length of the vowel part of the unit immediately before the nasal sound (including the cursive sound /N/) is longer than the predetermined set duration length, the above-mentioned The value of the vowel part of the immediately preceding unit is cut to the set duration length, and conversely, if the duration length of the vowel part is shorter than the set duration length, the value representative of the vowel part of the above unit is cut to the set duration length. 1. A method for synthesizing speech rules, characterized in that the method is configured to perform a process of filling in gaps at the rear so as to match a duration length. 2. In the process when the above-mentioned nasal sound is determined, the value of the parameter representing the amplitude of the above-mentioned nasal sound, the vowel immediately before the nasal sound, or both phonemes is corrected so that they are connected continuously at the boundary point of the two phonemes. A speech rule synthesis method according to claim 1, characterized in that: