JPS60225198A - Voice synthesizer by rule - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

[Detailed Description of the Invention] A) Industrial Application Field The present invention relates to a speech synthesis device based on rules,
By inputting any character string into the device, any synthesized speech can be obtained.

(Example) A speech synthesis device according to conventional technology rules stores speech parameters representing the characteristics of speech in units such as phonemes or syllables, and performs processing such as interpolation and length adjustment on the speech parameters according to the input character string. By connecting these, synthesized speech can be generated for any string of characters.

Conventionally, in this type of device, there are devices in which five tone parameters are determined in a one-to-one correspondence with the type of phoneme or syllable, and devices in which the phonetic parameters are determined in a one-to-many correspondence with the type of phoneme or syllable. That is, there are cases in which multiple types of speech parameters correspond to the same syllable. The latter is used for the same phoneme or syllable, for initial, middle, and
It stores a plurality of types of speech parameters, such as those for word endings, and selects speech parameters to be connected depending on the position of the phoneme or syllable in the word.

Furthermore, in order to further improve the naturalness of the synthesized speech, a speech synthesizer based on rules can specify the accent type of the word, etc.
A pisochi frequency is set, and a pisochi parameter is given to the audio parameter generated by the method described above. For example, for the same voice parameter, the accent of a word can be changed by changing the parameter representing the pitch frequency.

However, in a system that uses LSP, PARCOR, etc. as audio parameters, the parameter representing the vinyl frequency and other audio parameters are not completely independent. According to the Proceedings of the Acoustical Society of Japan (1-2-11, 19
(October 1981).

C) @Ming's Purpose The present invention has been made to solve the above problems, and provides a speech synthesis device that prevents deterioration of sound quality.

D) Structure of the Invention The speech synthesis device of the present invention is connected to a speech parameter storage section that stores speech parameters representing characteristics of speech in units such as phonemes or syllables, and reads out speech parameters from the storage section according to input character strings. In a speech synthesis device based on rules, which includes a speech parameter connection section that synthesizes speech based on the speech parameters generated in the speech parameter connection section, and a speech synthesis section that synthesizes speech based on the speech parameters generated in the speech parameter connection section, It stores a plurality of audio parameters representing , and selects the audio parameter based on the value of the bi-rich frequency.

E) Embodiment FIG. 1 does not show an embodiment of the speech synthesis device of the present invention. As shown in Figure 2, the voice memory selection team (1) has the following information:
The number of moras that roughly corresponds to the number of syllables to be synthesized and the voice parameters corresponding to each syllable for each combination of accent type are: high-pitched voice memory (2), middle-tone voice memory (3), and low-pitched voice memory. Out of (4), select "High", "Medium", which audio parameter in the audio memory should be connected.
Information indicated by "low" is stored. i.e. 4 of 4 syllables
The sounds represented by letters belong to four moras, and the accent types include those in which there is no accent on any of the syllables,
For example, ``1-omokage'' is type O, with an accent on the first syllable, for example, ``ennichi'' is type 1, and with an accent on the second syllable, for example, ``Uguisu'' is type 2, and the 6th syllable is type 2. Something with an accent on the syllable, such as "bangou"
is classified as type 6, and the accent positions and the pitch, j, and chi frequencies are relatively high, so high pitch voice parameters are set for this accent position, and the pitch is set for the last syllable. Since the ηchi frequency is relatively low, bass audio parameters are set. To explain Fig. 2 here, it schematically shows the speech memory selection table (1), and it shows which number of moras M and which speech parameters correspond to each syllable for each accented saint. "High", "Medium", "Low" using audio parameters
The information is stored as shown in . Furthermore, in Figure 1, (5) is a keyboard with character keys arranged, (6)
is a decoder that receives key operation signals from the keyboard (5) and converts them into character signals in syllable units corresponding to the keys.

(7) is the address where the character signal from the decoder (6) and the synthesis parameters in the treble voice memory (2), medium tone voice memo (January 3), and bass voice memory (4) exist. The audio parameter address table associated with (
8) is the audio parameter address table (7) II)
Voice memory group +21, (3+,
+41 address counter.

On the other hand, (9) is an address counter of the voice memory selection table (1) determined based on the mora number and accent type of the synthesized voice, and is the address counter of the voice parameter readout control circuit (1).
01 according to the output of the audio memory selection table (1),
The audio parameters are read from the address indicated by the address counter (8). (!1) is the parameter area (111)
This is an audio data buffer memory consisting of a parameter area (111) and a bi-rich storage area (112), and new audio parameters are sequentially stored in response to key input from the board (5). . In addition, the pitch pattern memory 112) stores parameter values indicating the number of moras and the number of pitch patterns corresponding to the accent type, and the value of the parameter corresponding to the synthesized voicing can be read out. and stored in the Pisochi area (112) in a time-expanded form. (13) is a counter that counts the number of moras corresponding to the teacher's number of key operation signals on the keyboard (51), and (14) is an accent type designation switch that specifies the accent type of the TS voice. (15) is the combination signal of the mora number obtained from the counter (13) and the accent type (specified in 141) in the monophonic parameter selection table (]). Address counter (9) and address counter (10) of pitch pattern memory 0z. (171 is a speech synthesis circuit, and the above-mentioned speech data reference memory 01)
An audio signal is synthesized and output based on the audio parameters stored in the audio signal, and this synthesized audio is generated from the speaker Q9) via the amplifier.

Next, we will explain the operation using the case of speech synthesis for "Japanese warbler" as an example. But before that, we analyze the natural speech "Japanese warbler" and find that it is a 4-mora type 2, and its bi-rich pattern. is as shown in FIG. According to the figure, if the threshold S1 for bass and middle tones is set to 200H1, and the threshold S2 for middle and treble tones is set to 300Hz, "uJ" is a middle tone, "gu" is a high tone, You can see that "i" is a middle tone and "su" is a low tone.

Then, "u", "gu", "i" from the keyboard (5)
, the character string "su" is input. At this time, counter OJ
The voice to be synthesized is set to 4 moras by counting the number of times the key operation signal is generated. Also, the operator sets the accent type to type 2 using the accent type command switch...J Chiro 4). Therefore, the signal 4 mora type 2 is sent from the Pi-Sochi pattern designation circuit 05) to the address counter (9), and the address of the audio memory selection table (1) specified by the address counter (9) contains the 4 mora type 2 signal. As a current table, a table showing "medium", "high", "medium", and "low" as shown in Fig. 3 is stored to match the actual thickness of the Pi-Sochi pattern in Fig. 5. ing.

On the other hand, the character ``u'' input from the keyboard (51) is converted into a character signal by the decoder (6) 1, and the syllable ``ノ'' is determined by the voice parameter address table (7) as shown in Figure 4. Since the five-tone parameter representing the characteristics of Ro is present at address 10, a value of 300 is set in the address counter (8).

Audio parameter readout i+i control circuit (IO+ is 11
3 of the mid-tone voice memory (3) corresponding to the first phrase of the 4-mora type 2 specified by the voice memory selection table (1)
Set the address 0000 audio parameters to 14. Parameter area (11) of Jinsei data base/sofa memory (11)
1).

Similarly, in response to the character signal of the following character "gu", the voice parameter representing the characteristics of the syllable "gu" is read out from the treble memory (2), and the parameter area (11
1), and in the same way, audio parameters for all characters are stored.

On the other hand, the output of the 4-mora type 2 bit pattern designation circuit (the old one was sent to the pitch pattern memory a sudden address counter DI; 112).

The voice parameters stored in the voice data buffer memory 01) through the above processing are sent to the voice synthesis circuit (171), and are generated from the speaker 091 via the amplifier (+8) as synthesized voice.

f) Effects of the Invention As is clear from the above explanation, when the speech synthesis device according to the rules of the present invention synthesizes words with various pitch frequencies, it is possible to synthesize the phoneme or syllable in the word by the value of the parameter indicating the pitch frequency. Since the optimum voice parameters are selected and used as the voice parameters representing the characteristics, deterioration in sound quality due to differences in pitch frequencies can be reduced, and high quality synthesized speech can be obtained.

[Brief explanation of drawings]

FIG. 1 is a block diagram of an embodiment of the speech synthesis device of the present invention, @
Figure 2 is a schematic diagram of the audio memory selection table, ff13[M
is a partial schematic diagram of a 4-mora type 2 voice memory selection table, and FIG. 4 is a schematic diagram of a voice parameter address table.
FIG. 5 is a curve diagram of the bi, , and c frequencies of the 4-molar type 2 word ``ukuisu''. +21, (31, (4)... voice memory, (51...
・Keyboard, 01)...Audio data buffer memory, (12)...Pi-S+pattern memory, (15)
)...Pi9chi pattern designation circuit, (1'TI...speech synthesis circuit, 09...speaker. Applicant Sanyo 1!1m Co., Ltd. Agent Patent attorney Shizuo Sano

Claims (1)

[Claims] a speech parameter storage section that stores speech parameters representing the characteristics of speech in units such as phonemes or syllables; a speech parameter connection section that reads and connects the speech parameters from the storage section according to an input character string; and the speech parameter connection section. In the speech synthesis device according to the rules, the speech parameter storage section is configured to store speech parameters representing characteristics of the speech of units such as the same phoneme or syllable. 1. A speech synthesis device based on rules, characterized in that a plurality of speech parameters are stored in the storage section, and the speech parameters in the storage section are selectively used according to the values of the B, C, and C frequencies.