JPH038000A - Voice rule synthesizing device - Google Patents

Abstract

PURPOSE:To reduce the cost by converting rhythm information which is supplied from outside into rhythm information corresponding to a synthesized voice speaker specified with speaker information. CONSTITUTION:Synthesis unit storage parts 5 - 7 are stored with relatively short acoustic parameters as synthesis units and also stored with synthesis units corresponding to synthesized voice speakers. A synthesis rule storage part 8 is stored with a synthesis rule for combining the synthesis units. A rhythm information conversion part 9 converts the rhythm information which is supplied from outside into the rhythm information corresponding to the synthesized voice speaker specified with the speaker information. A synthesis unit selection part 4 extracts synthesis units specified with the speaker information from storage parts 5 - 7 selectively. A voice parameter generation part 1 generates a voice parameter corresponding to the synthesized voice speaker according to the rhythm information and phoneme information converted by the conversion part 9 and the synthesis units and synthesis rule selected by the selection part 4. A voice synthesizer 2 synthesize a voice according to the voice parameter.

Description

[Detailed Description of the Invention] [Field of Industrial Application] The present invention relates to a speech rule synthesis device that creates synthesized speech by applying synthesis rules to synthesis units composed of relatively short acoustic parameters. .

(Prior art 1) A conventional speech rule synthesis device of this type is shown in FIG. 4. In this speech rule synthesis device, phonemes, CV,
A synthesis unit storage unit 5 uses relatively short acoustic parameters such as VC and VCV (C is a consonant and ■ is a vowel) as a synthesis unit.
Phonological information such as characters and phonetic symbols, and prosodic information such as length and intonation are inputted from the outside, and the synthesis rules stored in the synthesis rule storage unit 8 along with these information are used. A speech parameter generation section 1 generates speech parameters by combining synthesis units, and the speech parameters are supplied to a speech synthesizer 2 to produce synthesized speech. Using this method, it is possible to synthesize any natural sentence based on the given phonetic information and prosody information. By the way, although the above-mentioned prosody information can be generated according to rules from only the phonetic information, the synthesized speech synthesized according to the rules based on the prosody information generated in this way is insufficiently natural. Therefore, the special application for 1986-13
As shown in No. 8517 ([Speech code generation method]), if prosody information is extracted from natural speech, it is possible to generate highly natural synthesized speech.

[Problems to be Solved by the Invention] Incidentally, in order to create synthesized speech of different synthesized speech speakers, a speech rule synthesis device shown in FIG. 5 is used. This speech rule synthesis device includes synthesis unit storage units 5 to 7 which input speaker information specifying a synthesized speech speaker from the outside and store synthesis units corresponding to each synthesized speech speaker. a synthesis unit selection unit 4 that selectively extracts synthesis units specified by the speaker information from the synthesis unit storage units 5 to 7 corresponding to the synthesized speech speaker; It is now possible to create synthetic speech.

However, when creating synthesized speech for different synthesized speech speakers in this way, information regarding the length and strength of the synthesis unit in the prosodic information can be used in common for different synthesized speech speakers. However, information regarding height, that is, pitch information cannot be commonly used because it causes deterioration in the sound quality of synthesized speech. In other words, when synthesizing synthesized speech of different synthesized speech speakers even if the phonetic information is the same, it is necessary to input prosodic information corresponding to each synthesized speech speaker into the speech synthesizer. However, when extracting prosodic information from natural speech, it is necessary to modify the prosodic information while listening to the synthesized speech in order to ensure the naturalness of the synthesized speech.
The productivity was low and there were time and economic problems.

The present invention has been made in view of the above points, and its purpose is to provide speech rule synthesis that can use the same prosodic information even when creating synthesized speech of different synthesized speech speakers. The goal is to provide equipment.

[Means for Solving the Problems] In order to achieve the above object, the present invention externally provides speaker information for specifying a synthesized speech speaker, and provides information on the synthesized speech speaker identified by the speaker information. A prosodic information converter is provided for converting prosodic information given from the outside into prosodic information. Note that the above speaker information may be provided from the outside in association with phoneme information and prosody information. In addition, when specifying a male voice and a female voice using speaker information, when a male voice is specified using the speaker information in the prosodic information converter, it is 7α higher than when a female voice is specified.
All you have to do is convert the pitch data one octave lower.

[Operation] As described above, the present invention converts prosodic information given from the outside into prosodic information corresponding to a synthesized speech speaker not specified by the speaker information in the prosodic information converter, thereby converting prosodic information into different synthesized speech stories. Even when generating human speech, the same prosodic information can be input from the outside, reducing the number of times the prosodic information is created, which is the most time-consuming and expensive process when creating input information.
This makes it possible to reduce the input information creation cost per one synthesized sound door.

[Example 1] An embodiment of the present invention is shown in FIGS. 1 and 2. FIG.

In this embodiment, a case will be described in which synthesized speech of three synthesized speech speakers is generated using a PARCOR type speech synthesizer. This speech rule synthesis device also receives phonological information and prosody information for setting the tone quality of the synthesized speech and speaker information for specifying the speaker of the synthesized speech from the outside, and generates the synthesized speech based on these information. The speech rule synthesis device stores relatively short acoustic parameters as synthesis units, and also includes three synthesis unit storage units 5 to 7 that store synthesis units corresponding to each synthesized speech speaker, and synthesis units. Synthesis rule storage unit 8 contains synthesis rules related to how to connect synthesis units when combining, how to add intonation, etc., and prosodic information corresponding to the synthesized speech speaker specified by speaker information is stored from outside. a prosody information converter 9 that converts given prosody information; a synthesis unit selector 4 that selectively extracts synthesis units specified by speaker information from synthesis unit storage units 5 to 7; and the above-mentioned prosody information conversion section. Speech parameter generation that generates speech parameters corresponding to the synthesized speech speaker from the prosodic information and phonological information converted in step 9, the synthesis unit corresponding to the synthesized speech speaker selected by the synthesis unit selection section, and the synthesis rule. 1, a PARCOR-based speech synthesizer 2 that creates synthesized speech according to the speech parameters generated by the speech parameter generation section 1, and an acoustic pH device such as a D/A converter, a low-pass filter, and a speaker. It consists of an output processing section 3. This speech rule synthesis device converts an input character sequence as phoneme information into a CV tone sequence sequence (for example, / yokohama / (Yokohama)).
is segmented into four phoneme chain sequences: yo, oko, oha, and ama). In other words, this speech rule synthesis device employs the VCVM set synthesis method. The following prosodic elements are used in such a VCV editing and synthesis method.

1. The value of the pitch (point pitch) at the center of vowel gravity 2. The VCV chain length that gives the time interval between adjacent point pitches 3. The amplitude value at the point pitch position Therefore, in this speech rule synthesis device, the value of the pitch (point pitch) at the vowel gravity center point is given as 3. As shown in Figure 2, the three prosodic elements mentioned above are provided externally. In addition, the synthesis unit storage unit 5 stores ■C■ phoneme chain data as a synthesis unit, and this VC■
Phonological chain data is PARC which carries spectral envelope information.
It is expressed by an OR coefficient, a voiced/asexual coefficient indicating the nature of the sound source, amplitude information, etc. Furthermore, as a synthesis rule, ■ linearly connect the vowel constant parts according to the CV series, and linearly connect the point pitches, and from the amplitude level information at the point pitch, the amplitude level of the vowel constant part, the vowel transient part, and the consonant. The method is used to generate the amplitude level of the section.

Also, the above speaker information is 8! for synthetic voice speaker selection. It is configured to input from a mechanical switch, and when generating synthesized speech of one synthesized speech speaker, the synthesis unit storage unit 5
Only any of the synthetic units .about.7 are used.

Now, when the phonetic information, prosody information, and speaker information regarding a group of sentences are input, the synthesis unit selection section 4 selects one of the synthesis unit storage sections 5 to 7. In the following explanation, synthesized voice speaker AI: corresponding synthesis unit storage unit 5,
The case where is selected will be explained. When synthesized speech speaker A is selected in this way, the prosody information is converted to the prosody information converter 9.
The synthesized speech is converted into prosodic information corresponding to the synthesized speech speaker A specified by the speaker information. For example, how to convert this prosodic information! @2 As shown in Figure 2, the point pitch data is connected to the multiplier 10 according to the speaker information, and the multiplier 10
There are methods such as multiplying the point pitch data by a constant. Next, the speech parameter generation section 1 reads the necessary synthesis unit from the synthesis unit storage section 5 according to the phonetic information, and according to the converted prosody information and the synthesis rule stored in the synthesis rule storage section 8, the length, Generate voice parameters by adding intonation, etc. This audio parameter is
OR coefficients, pitch parameters, etc. are arranged in chronological order. These voice parameters are converted into a synthesized voice signal, which is a digital signal, by the voice synthesizer 2, and then sent to the output processing unit 3.
is output as a synthesized voice. In this way, in this embodiment, when generating synthesized speech having the same prosodic information for a plurality of synthetic speech speakers, the same phonological information and prosodic information can be used, and the storage medium that stores this information can be used. Alternatively, the amount of information to be transmitted can be reduced. Furthermore, if the above information is embedded in a barcode, for example, the user is freed from the complexity of selecting the location where the information is stored.

Furthermore, even when adding a synthesis unit storage section, only the prosodic information conversion section 9 needs to be changed without changing the phonological information or prosody information.
It will be placed in c position.

By the way, in the above embodiment, the speaker information is input from a mechanical switch for selecting a synthesized voice speaker, but even if the speaker information is stored in the same storage medium as the phonological information store V prosodic information, good. In this case, speaker information can be set in advance and given externally in association with phonetic gt information and prosodic information, eliminating the need for manual operations such as switching speakers while the synthesized speech is being uttered. Even when the speaker changes within a relatively short period of time, such as in a sentence, the speaker can be changed smoothly.

[Embodiment 2] This embodiment relates to a speech rule synthesis device that outputs a male voice and a female voice based on the same phoneme information and prosody information. In the case of this embodiment, at least two voices corresponding to male and female voices are used.
It is sufficient to provide two synthesis unit storage units 5 and 6, and specify male and female voices based on speaker information. In addition, when outputting a male voice and a female voice based on the same phonetic information and prosody information, the prosodic information converting section 9 has 172 as shown in FIG.
A circuit 11 may be provided, and the 1/2 circuit 11 may convert point pitch data in the prosodic information one octave lower than in the case of a female voice when a male voice is specified by the speaker information. In this way, the point pitch data, which was conventionally lowered appropriately, can be lowered by exactly one octave, which can be effectively used in, for example, a voice synthesizer singing device. Moreover, there is an advantage that the process of reducing the point pitch data to 1/2 can be easily performed.

In addition, in the description of each of the above-mentioned embodiments, the audio parameters are expressed as PARCOR, which is a linear predictive analysis parameter.
coefficients, and three types of synthesized speech speakers.
It is not limited to the above. Furthermore, although the synthetic unit also uses ■C■ chain data, it is not limited thereto.

[Effect of the Invention 1] As described above, the present invention converts prosodic information given from the outside into prosodic information corresponding to the synthesized speech speaker specified by the speaker information in the prosodic information converter, so that different Even when generating the voice of a synthetic voice speaker, the same prosodic information can be input from the outside, which reduces the number of times the prosodic information is created, which is the most time-consuming and expensive process when creating input information. This has the effect of lowering the cost of creating input information.

[Brief explanation of the drawing]

FIG. 1 is a schematic block diagram of an embodiment of the present invention, FIG. 2 is a block diagram of the same prosody information converter as above, FIG. 3 is a block diagram of the prosody information converter of another embodiment, and FIG. FIG. 5 is a schematic diagram of another conventional example. 1 is a speech parameter generation section, 2 is a speech synthesizer, 4 is a synthesis unit selection section, 5 to 7 are synthesis unit storage sections, 8 is a synthesis rule storage section, and 9 is a prosodic information conversion section.

Claims (3)

[Claims] (1) A speech rule synthesis device that receives phonological information and prosody information that sets the sound quality of synthesized speech and speaker information that specifies the speaker of synthesized speech from the outside, and generates synthesized speech based on these information. a synthesis unit storage section that stores relatively short acoustic parameters as synthesis units, and also stores synthesis units corresponding to each synthesized speech speaker; and a synthesis rule that stores synthesis rules for combining synthesis units. a storage unit, a synthesized speech specified by the speaker information; a prosodic information conversion unit that converts prosodic information given from the outside into prosodic information according to the speaker; and a synthesis unit memory for the synthesis unit specified by the speaker information. a synthesis unit selection section that selectively extracts the prosodic information from the prosodic information conversion section, the prosodic information and phonetic information converted by the prosodic information conversion section, synthesis units corresponding to the synthesized speech speaker selected by the synthesis unit selection section, and synthesis rules. A voice rule synthesis device comprising: a voice parameter generation unit that generates voice parameters corresponding to the synthesized voice speaker; and a voice synthesizer that creates synthesized voice according to the voice parameters generated by the voice parameter generation unit. . (2) The speech rule synthesis device according to claim 1, wherein the speaker information is provided from the outside in association with phonological information and prosody information. (3) Identify male and female voices using the above speaker information, and
2. The speech rule synthesis device according to claim 1, wherein when said prosodic information converter identifies a male voice based on speaker information, the point pitch data is converted to be one octave lower than that for a female voice.