JPH0833751B2 - Speech synthesis method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION A. Field of Industrial Application The present invention relates to a speech synthesis method by rule synthesis using an acoustic tube.

B. Summary of the Invention In the present invention, in the speech synthesis method by rule synthesis, when the syllable data is given to the acoustic tube to change the cross-sectional area, the cross-sectional area is controlled to precede the energy of the syllable data. This makes it possible to easily synthesize sounds similar to human utterances and to improve the naturalness of speech.

C. Conventional technology Electronic devices that synthesize and output speech artificially have recently been low-priced by one or several chips of speech recognition and speech synthesis LSIs due to speech information processing and semiconductor large-scale integrated circuit technology. , And various methods have been proposed depending on the purpose of use and constraints. For this speech synthesis,
Record the raw voice generated by humans, combine it appropriately and edit it into sentences, and a recording / editing method that does not directly use the human voice, but extracts only human voice parameters. , There is a method of artificially producing a voice signal by controlling the parameter in the voice synthesis process.

There is a percall (PARCOR) method which is widely used because a high quality synthesized sound can be obtained by this parameter method.

When a voice is handled by an electronic computer, the voice waveform is sampled at a certain cycle, the value of the voice signal at each sampling point is converted from analog to digital, and the value is displayed with a code of 0 and 1. It is necessary to increase the number of bits in order to record faithfully to the analog signal, but the voice synthesis signal requires a very large amount of memory.

Therefore, various highly efficient coding methods have been researched and developed in order to reduce the amount of this information as much as possible.

As one of the methods, there is a delta modulation method in which the information of one audio signal is at least 1 bit. This method uses one bit to judge whether the next audio signal value is higher or lower than the current value, and if the value is higher, the code "1" is given, and if the value is lower, the code "0" is given. In the actual system configuration, a fixed amplitude step amount (delta) is set, and the audio value obtained by the encoding up to now and the input audio signal are set so that errors are not accumulated. The residual signal of is encoded.

Such a configuration is called predictive coding, and uses a linear prediction method (predicting from several previous sample values) and a Percoll method (a partial autocorrelation function called a Percoll coefficient k instead of the prediction coefficient of the linear prediction method). ).

D. Problems to be Solved by the Invention As described above, the method using predictive coding has a problem that articulatory coupling corresponding to a joint between sounds is difficult. For example, in the utterance from a vowel to a consonant to a vowel, the vowel-vowel joint sound is cut off in the process of a vowel's steady state to a consonant to a consonant, and a vowel transition to a vowel's steady tone. Does not give a natural feel when played.

E. Means for Solving Problems This invention processes an input character by a Japanese processing unit using a language registered in a dictionary registration unit provided in advance, and then processes the processed language. The syllable processing section divides the pitch, energy, and duration into syllabic data, and when this syllable data is given to the acoustic tube to change the cross-sectional area, the cross-sectional area of the acoustic tube is controlled ahead of the energy for a certain period of time. To obtain a voice synthesis output.

F. Action When outputting the language processed by the Japanese processor as speech, the language is first processed by syllables. By this processing, syllable data such as pitch and energy is obtained. When this syllable data is given to the acoustic tube, if the cross-sectional area is controlled to precede the energy by a certain amount of time and the voice synthesis output is sent, it is possible to synthesize a sound similar to human speech and to improve the naturalness. You can

G. Example First, the prior invention (the Japanese Patent Application No. 62-917) that forms the basis of this invention.
No. 05) will be outlined. The invention of the prior application is configured as follows.

Since human sound is generated by changes in the shape of the human mouth and the length and cross-sectional area of the acoustic tube, the traveling wave phenomenon that represents the transmission of sound waves in these acoustic tubes is analyzed by an acoustic equivalent circuit to determine the disconnection of the acoustic tube. Focusing on the fact that the area is inversely proportional to the surge impedance, changing the surge impedance changes the cross-sectional area in a simulated manner, and continuously changing the surge impedance enables smooth articulation coupling. This is to facilitate the synthesis of sounds similar to the utterance of and to improve the naturalness of the sounds.

In the invention of the prior application configured as described above, the cross-sectional area change of the vocal tract at the time of voice generation is, for example, in the case of the utterance of "A", the exhaled breath extruded from the lungs with the throat narrow and the lips open. The vocal cords open and close the breath intermittently in the vocal tract (acoustic tube)
The sound wave that is repeatedly reflected in the sound comes out as a sound waveform of "A". When "a" has a wide throat and narrow lip, the voice waveform of "a" is output.

In this way, the frequency is determined by the mouth preference, and if the mouth preference is simulated, "a" or "a" is uttered. The mouth shape can be simulated by the cross-sectional area of the acoustic tube, and the change of the cross-sectional area of the acoustic tube can be simulated by the change of surge admittance. Therefore,
If the surge admittance is changed, the mouth appearance can be simulated. Since the change of the surge admittance can be changed very easily on the electric circuit, various voices can be synthesized by the electric signal. 4 (A) is obtained by simulating the vocal tract by connecting the sound tube having a cross sectional area different from the cross-sectional area _{A 1, A 2 ... A n} . In the figure (a), the acoustic impedance is replaced with an LC circuit of an electric circuit.
Each acoustic tube is one LC line, and the whole is n-1 of the concentrated line
The electric circuit of. Further, FIG. 4C is a traveling wave equivalent model diagram, in which the acoustic impedance Z ₁ ,
Z ₂ ... Z _n is inversely proportional to the cross-sectional area of the acoustic tube (acoustic admittance) and proportional to the velocity C of the sound wave (ρ is the air density), Becomes In the figure, Zg is the sound source impedance, Z _L is the radiation impedance, and the arrows between the blocks are
It represents a traveling wave and a backward wave.

When uttering the voice "A" now, by giving the appearance of the mouth "A" at the cross-sectional area of the acoustic tube corresponding to the tip of the lip and intermittently applying the impulse P,
When the sound of "A" is obtained and the sound of "A" is produced from "A", the cross-sectional area of the acoustic tube corresponding to the tip of the lip is narrowed to give the appearance of "A". I "is obtained.

When impulses P are given continuously and intermittently and the entire cross-sectional area is changed arbitrarily in the mouth of "a", the vocal tract is simulated by n acoustic tubes shown in FIG. Move each cross section from "A" to change the mouth appearance to "A"
Will be changed continuously. Changing the cross-sectional area of this acoustic tube is performed by gradually changing the surge impedance.

Therefore, since the cross-sectional area can be changed continuously, it is of course possible to obtain steady-state sounds of "a" and "a".
Further, since the impedance can be continuously changed, it is possible to obtain an intermediate sound, that is, a sound between sounds. Therefore, there is no break in the sound, and articulatory coupling similar to human pronunciation is smoothly performed.

Next, considering the propagation velocity of sound waves, this is similar to a transient phenomenon when an impulse is applied to an electric line having a length of l and having an LC.

That is, the line having LC as shown in FIG. 5 is equivalently expressed as shown in FIG. Here, the surge impedances Z ₀₁ and Z ₀₂ seen from both ends are Becomes

Considering the traveling wave that arrives from the other party as an equivalent current source, If there are n delay circuit blocks Z in the middle, the next current is output after n hours. That is, what has occurred in the circuit on the left has arrived on the right after τ.

I ₂ is the current on the feed tube side Becomes However, in the digital calculation, since the voltage or current is subdivided, V ₁ and V ₂ indicate the voltage at the measurement time t, and τ indicates the elapsed time.

In Fig. 6, if impulses are applied to the L and C circuits, τ
It goes out to the output tube side after time. And, it is equivalently expressed that what has arrived τ time ago has also reached the other party. Setting the line length 1 to 1 facilitates calculation by normalizing the delay block n to 1. l
In case of dividing into 3 cm, the delay block n may be set to 3 blocks.

Figure 4 (a) shows that the human vocal tract is about 17 cm for men, so 1c
If simulated with 17 acoustic tubes in m steps, the waveform entered from A ₁ will come out from the _An side in 170 μsec if the half-cycle current is divided into 10 and Δt is 10 μsec.

Therefore, the arithmetic processing unit performs arithmetic processing corresponding to the cross-sectional area changes of the acoustic tube cross-sections A _{1 to} A _n , and has a current value and a function of each part flowing through each equivalent circuit of the acoustic tubes A _{1 to} A _n as a table. If the arithmetic processing is performed, the memory, the arithmetic means for calculating the current value of each part of the equivalent circuit, and the arithmetic means for calculating the current value using the current value of the equivalent circuit adjacent to the equivalent circuit are provided. An audio signal is obtained and its output is D /
If A-converted and output to the speaker, it will be output as sound from the speaker.

Next, a description will be given of an example of synthesizing voice by a rule from a character input signal using the above acoustic tube model.

FIG. 1 is a block diagram for explaining an embodiment of the present invention. Reference numeral 1 is a Japanese language processing unit, which receives a sentence written with a mixture of kanji and kana as an input and stores it in a dictionary registration unit. Corresponding to the dictionary registration of 2, natural language analysis of clauses, phrases, sentence delimiters, morpheme classification, and further accent processing, phonetic conversion of this, and addition of intonation to create sentence processing data. is there. Reference numeral 3 is a syllable processing unit for processing syllable parameters from the data processed by the Japanese processing unit 1. The syllable parameters processed by the syllable processing unit 3 are 110 to 140 consonants (a word that can be normally spoken is Pitch (pitch P), sound intensity (energy E) and duration (T) for each syllable of about 110 syllables
It is divided into. An example of this syllable parameter is shown in FIG. This Fig. 2 shows an example of the case of "Sakura".
Pitch P, energy for each syllable of “A”, “KU”, and “RA”
E, time T is shown.

The syllable parameters obtained as described above are given to the acoustic tube model. At this time, the cross-sectional area control unit 4, the pitch control unit 5, and the energy control unit 6 are controlled as follows. The cross-sectional area control unit 4 inserts a delay timer 7 in the preceding stage of the energy control unit 6 so that the energy control unit 6 can be controlled ahead of the energy control unit 6 by a certain time, and delays the energy for a certain time. That is, as shown in FIG. 3, the cross-sectional area always rises earlier than energy by τ time. By controlling in this way, the utterance is the same as when a human utters, and the voice becomes natural. This is because when humans speak, they have the same control as when their mouth first opens and then energy is applied.

The outputs of the cross-sectional area control unit 4, the pitch control unit 5, and the energy control unit 6 are D / A converted by a D / A conversion unit 8 and a voice output is sent from a speaker (not shown).

H. Effects of the Invention As described above, according to the present invention, when a human utters, the cross-sectional area is controlled to precede the energy so that the same control as that of making a mouth and then making a sound is performed. This makes it easy to synthesize sounds similar to human speech,
It is possible to improve the naturalness of voice.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram showing an embodiment of the present invention, FIG. 2 is an explanatory diagram of syllable parameters, FIG. 3 is an explanatory diagram of a relation between cross-sectional area and energy, and FIG. 4 (a). Is an acoustic tube model diagram, FIG. 4 (a) is an electrical circuit model diagram of the acoustic tube, FIG. 4 (c) is a traveling wave equivalent model diagram of the acoustic tube, and FIG. The circuit diagram and FIG. 6 are equivalent circuit diagrams of FIG. 1 ... Japanese processing unit, 2 ... dictionary registration unit, 3 ... syllable processing unit, 4 ... section area control unit, 5 ... pitch control unit, 6 ...
… Energy controller, 7 …… Delay timer.

Claims (1)

[Claims] 1. An input character is processed by a Japanese language processing unit using a language registered in a dictionary registration unit provided in advance, and then the processed language is processed by a syllable processing unit in terms of pitch, energy, When syllable data is created by dividing into durations, and when this syllable data is given to the acoustic tube to change the cross-sectional area, the cross-sectional area of the acoustic tube is controlled ahead of the energy for a certain period of time to obtain a speech synthesis output. A speech synthesis method characterized by the above.