JP2894447B2 - Speech synthesizer using complex speech units - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech synthesis device using a complex speech unit, and more particularly to a speech synthesis device that performs speech synthesis by editing, combining, and transforming a speech unit. [Prior Art] Various speech units have been proposed in a conventional edit-type speech synthesis method and a rule-based speech synthesis method. Such speech units include phoneme, CV (consonant, vowel) syllable, semisyllable (consonant, vowel and vowel / consonant concatenation), VCV (vowel / consonant / vowel) concatenation, CVC (consonant) , Vowels and consonants), CVCV (consonants, vowels, consonants and vowels), words, and phrases, etc. In the conventional speech synthesis system, any one of these types of speech units is used as a base, Even if the other one or two types of audio units are used as a supplement, at most 2-3
Speech was generated by editing, combining, and transforming different types of speech units. [Problems to be Solved by the Invention] FIG. 4 is a diagram comparing conventional various voice units into two types, one having a short unit length and the other having a long unit length, and comparing the superiority and demerit in the configuration of a voice synthesis system. It is. In FIG. 4, it is desirable that the storage capacity required for storing units that are a problem in configuring the speech synthesis system be small, and the shorter the length of the speech unit is, the more advantageous. In addition, the man-hour for creating the synthesis unit is more advantageous as the length of the voice unit is shorter. In terms of the availability and flexibility of unit duplication, the shorter the voice unit, the larger the sound unit. On the other hand, in the unit transformation method, the unit connection method, and the like that affect the voice quality, the longer the length of the voice unit, the less the problem can be. As is clear from the above comparison, it has been extremely difficult to configure a speech synthesis system with only two types of speech units in a well-balanced manner. Therefore, a main object of the present invention is to provide a composite speech unit that can obtain more efficient speech synthesis by using various types of speech units having various structures and partial units included therein. An object of the present invention is to provide a speech synthesizer used. [Means for Solving the Problems] The present invention relates to a speech synthesizer using a complex speech unit for performing speech synthesis by editing speech units or for synthesizing an arbitrary vocabulary according to rules. , A morpheme, a word, a phrase, and a speech unit file that stores in advance a large number of speech units having various lengths and structures up to sentences, and a phonological environment and prosodic information in which partial units of the phonological unit of the synthesis location are placed. I want to output a speech control file that stores in advance an extraction environment for selecting a part of a speech unit, a rule that gives specific conditions such as a fundamental frequency of a phoneme, a phoneme time length, and tables that are parameters for applying the rule. When a prosody control signal indicating a prosody including the pitch, speed, and loudness of the voice corresponding to the voice content is input, rules and various rules given from the voice control file are input. Based on the table, a speech unit selection means for selecting a speech unit sequence suitable for synthesis from the speech unit file, a combination unit for combining the selected speech unit sequence, and based on the combined synthesis unit sequence, And a composite waveform generating means for generating a composite waveform. [Operation] The speech synthesizer using the composite speech unit according to the present invention is suitable for synthesis from a speech unit file based on rules and tables given from the speech control file in response to input of a prosody control signal. The synthesized speech unit sequence is selected and synthesized, and a synthesized waveform is generated based on the synthesized synthesis unit sequence to perform speech synthesis. Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. FIG. 1 is a schematic block diagram of a speech synthesizer according to the present invention. In FIG. 1, a prosody control signal such as a phoneme sequence signal, an accent, and breathing corresponding to a speech content to be output is input to an input terminal 1. The prosody control signal input to the input terminal 1 is supplied to the voice unit selection unit 2. The voice unit selection unit 2 selects a voice unit string suitable for synthesis from a large number of voice units having various structures stored in the voice unit file 3 based on the input information. The speech control knowledge file 4 stores rules and tables for giving specific synthesis conditions such as an extraction environment, a fundamental frequency, and a phoneme time length. When selecting a speech unit sequence, the speech unit selection unit 2 extracts a desired synthesis unit extraction environment, a fundamental frequency, and a phoneme time length from the speech control knowledge file 4 based on the phoneme environment and the prosody information of the synthesis location. Specific composition conditions such as are given, and these synthesis conditions are used as criteria for unit selection. The voice unit sequence selected by the voice unit selection unit 2 is provided to the unit combining unit 5, and the selected units are combined. The combined speech unit is provided to a combined waveform generation unit 6, and a combined waveform is generated based on the combined unit sequence obtained by the unit combining unit 5. Then, the generated sound is output to the output terminal 7. Further, the method according to the present invention is characterized in that various types of units having various structures are used, and the acoustic parameters expressing the voice units used when implementing the method according to the present invention and various types used for unit selection are used. For the standard of speech, speech control knowledge, and synthesizer for generating speech waveforms,
It does not specify this at all and is applicable to all. FIG. 2 is a diagram showing a specific example of a speech unit file used in the present invention. The top 50 voice units are shown as a sound unit set including the subset. In FIG. 2, the number to the right of each phoneme connection indicates the appearance frequency, and the numbers are arranged in descending order of the frequency for each phoneme connection number. Further, in FIG. 2, 108 phoneme connections connected by a circle constitute a unit file, and the partial connection includes the top 50 of these 1 to 4 connections. More specifically, the phoneme “u” having the number of concatenation 1 is included in the phoneme “ru” having the number of concatenation 2, and this phoneme “ru” is included in the phoneme “eru” having the number of concatenation 3; This phoneme “eru” is included in the phoneme “keru” having four concatenations. Therefore,
The phonemes "u", "ru", and "eru" do not need to be stored in the voice unit file, and can be obtained by extracting the phonemes "keru" as partial units. FIG. 3 is a diagram showing an example of a voice unit sequence selected when the example shown in FIG. 2 is used as a voice unit set and, for example, "sakura blooms" / sakuragasaku / is input. As shown in FIG. 3, / sa / at the beginning of the sentence is given by a subsequence generated from four concatenations / saku /, and / kura / is four concatenations / k
Given by ura /, a voice unit sequence of "Sakura" is realized. Furthermore, "ga" is realized by / ga / of four connections / gaku /, and "blooming" is similarly realized by four connections / saku /. [Effects of the Invention] As described above, according to the present invention, the diversification of audio units allows free choice of storage capacity allowed for unit storage, man-hours allowed for creation, etc. In addition to the advantage that the unit set can be defined efficiently, there is an advantage that the unit can be used effectively and flexibly, such as partial use of a short connection included in a long audio unit. further,
Since a longer unit is used, there is little deterioration in voice quality due to connection between units, and good voice quality can be expected.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a schematic block diagram for implementing the method of the present invention. FIG. 2 is a diagram showing a specific example of a sound unit file. FIG. 3 is a diagram showing an example of a speech unit sequence selected for an input phoneme sequence. FIG. 4 is a diagram showing the relationship between the speech synthesis unit length and problems in the speech synthesis system configuration. In the figure, 1 is an input terminal, 2 is a voice unit selection unit, 3 is a voice unit file, 4 is a voice control knowledge file, 5 is a unit combining unit, 6 is a synthesized waveform generation unit, and 7 is an output terminal.

Claims (1)

(57) [Claims] This is a speech synthesizer that uses a complex speech unit that synthesizes speech by editing speech units or synthesizes an arbitrary vocabulary according to rules. The speech synthesizer uses various sounds ranging from phoneme levels to syllables, morphemes, words, phrases, and sentences. A speech unit file that stores a number of speech units having the length and structure in advance, a phoneme environment in which the phoneme subunits of the synthesis location are placed, and an extraction environment that selects speech unit parts based on prosodic information. A voice control file that stores in advance rules that give specific conditions such as the fundamental frequency and phoneme time length of the voice and tables that are parameters for applying the rules, the pitch, speed, and loudness of the voice corresponding to the voice content that you want to output In response to the input of a prosody control signal indicating a prosody including, based on rules and tables given from the voice control file, Based on a speech unit selection unit that selects a speech unit sequence suitable for synthesis, a combination unit for combining the speech unit sequences selected by the speech unit selection unit, and a synthesis unit sequence combined by the combination unit. A speech synthesizer using a composite speech unit, comprising: a synthesized waveform generating unit for generating a synthesized waveform.