JPH01292400A - Speech synthesis system - Google Patents

Abstract

PURPOSE:To generate a natural synthetic voice while smoothing modulation coupling more by performing arithmetic interpolation in both a time-basis direction and a voice channel direction and simulating the motion of a tongue. CONSTITUTION:A parameter interpolation processing part 5 performs interpolation not only in the time-base direction, but also in the voice-channel direction. For the simulation of the motion of the tongue, the interpolation is carried out only within the range of an acoustic tube corresponding to positions of the voice channel where the tongue can move. Namely, the voice channel positions where the tongue can move are denoted as KI-KE and then 0<=KI<=KE<=n-1 (n: number of stages of the acoustic tube) holds; and sectional product interpolation is performed when 0<=i<=KI and KE<=i<=n-1 hold for a sectional product parameter Ai at an optional time. Then an arithmetic part 6 finds the digital value of a current at the same timing with the arithmetic interpolation and this value is A/D-converted 7 to generate a voice by a voicing part 8.

Description

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

B1 Summary of the Invention The present invention regards the human vocal tract as a group of acoustic tubes, and by associating this with a group of circuit elements for surge impedance components, it is possible to simulate speech based on the current wave at the output end of the group of circuit elements. In this method, auxiliary calculations are performed not only in the direction of the time axis but also in the direction of the vocal tract to simulate the movement of the tongue, further smoothing the articulatory combination and producing natural synthesized sounds.

C8 Conventional technology Electronic devices that artificially synthesize and output so-called sounds, such as voice synthesis and music synthesizers (electronic musical instruments), have recently been developed into LS for voice recognition and voice synthesis using one or several chips.
I can now be realized at low cost through audio information processing and semiconductor large-scale integrated circuit technology, and various systems have been proposed depending on the purpose of use and constraints. This speech synthesis method involves two methods: recording and editing raw human speech, which combines them appropriately and editing them into sentences; There is a parameter method that extracts only the parameters and controls them during the speech synthesis process to artificially create a speech signal.

In the parameter method, the audio waveform is sampled at certain intervals, the audio signal value at each sampling point is converted from analog to digital, and the values are displayed as codes of 0 and 1. In order to perform faithful recording, it is necessary to increase the number of bits, which requires a large memory capacity.

Therefore, various highly efficient encoding methods are being researched and developed in order to reduce the amount of information as much as possible.

One such method is a delta modulation method in which at least one bit corresponds to the information of one audio signal. In this method, one bit is used to determine whether the next audio signal value is higher or lower than the current value, and if it is higher, the sign @1 is used.
”, if it is low, the code “0” is given and the audio signal is encoded.In the actual system configuration, a constant amplitude step amount (delta) is determined, and the previous code is changed to prevent errors from accumulating. The residual signal between the audio value obtained by the encoding and the input audio signal is encoded.

This type of predictive coding is called the linear prediction method (predicting from several previous sample values) and the Percoll method (using a partial autocorrelation function called the LiniPercoll coefficient, which changes the prediction coefficient of the linear prediction method) There is.

D1 Problems to be Solved by the Invention Of the conventional speech synthesis methods, the recording and editing method has a problem in that the types of words and sentences that can be synthesized are limited.

In addition, in methods using predictive coding, it is difficult to create articulatory combinations that correspond to the joints between sounds, and a method for combining synthesis units has not been established. In the process of going from a steady state through a transition to a consonant, and then through a vowel transition to a steady vowel sound, the sound at the joint between vowels is cut off. Therefore, there is a problem that the sound lacks smoothness and does not give a natural feeling when heard by humans.

An object of the present invention is to provide a speech synthesis device that can synthesize sentences of any amount and provide a smooth sound that is close to actual human speech and can give a natural feeling to the listener. It's about doing.

Means and Actions for Solving the E0 Problem In order to radiate sound out of the mouth, a sound source is required, and this sound source is produced by the vocal cords. -The vocal cords have the function of intermittent exhalation by opening and closing two folds, and these intermittent intervals generate airflow called puffs, and when the vocal cords are tensed, tension is applied to these folds, causing the folds to open and close. The frequency increases, producing a high-frequency puff sound. When the exhalation flow is increased, the sound becomes louder.

When this sound source wave passes through a cylindrical acoustic tube like the vocal tract, a resonance phenomenon causes certain components of the sound waves from the open end to be emphasized and certain components to be attenuated, creating a complex vowel waveform.

Even if the sound source waves have the same waveform, the sound emitted from the mouth is affected by the shape of the vocal tract that the sound passes through before being emitted from the lips. That is, the sounds produced by humans are determined by the length and cross-sectional area of the vocal tract from the vocal cords to the lips, and the way the vocal cords vibrate.

The present invention has been made with this in mind, and considers the vocal tract as a group of acoustic tubes with variable cross-sectional areas.
Furthermore, the starting point is to realize the traveling wave phenomenon representing the transmission of sound waves such as acoustic waves using its equivalent circuit.

From the above, the vocal tract is represented by the equivalent model shown in Figure 2 (A) in which n one-dimensional acoustic tubes with cross-sectional areas A, ~A7 are connected in cascade, and each acoustic tube S1 ~ So is represented by the LC component only. When considered as a lossless distributed constant line T, ~T1, it is converted to the electric circuit shown in the same figure (a), and when an impulsive electric or current signal instead of a sound source is applied to this electric circuit, the output waveform becomes an audio waveform. It can be considered as the corresponding response output. and,
To change the articulation state of the vocal tract, the constants of each line T, ~T1 are required.

By changing C, the surge impedance can be changed.

In the equivalent circuit of the same figure (a), when an impulsive current is applied in place of the sound source, each line T.

~Tn are respectively surge impedances 2. It has a current source with an internal resistance of -2n, and is converted into an equivalent circuit with connection points of each line T, - T, and reflection and transmission.

This equivalent circuit is shown in the same figure. In the figure, E is the voltage source voltage corresponding to the sound source, Zo is its output impedance, ZL is the radiation impedance from the lips, and A is the forward wave at each connection point of the lines Tl to Tn. Current i. n"" I (W-11B is the backward wave current at each connection point, Z I-Z- is the cross-sectional area A, -A, respectively
, and air density ρ and sound velocity C, the surge impedance becomes ρC/A, ~ρC/A, ■ ~1111-11A and 118~Ins are the current source currents a appearing at the connection point of the lines T and ~T1. A"" a 111-11A and aha~
a0 is the shunt current of the current source.

These currents have the relationship shown in FIG. 2 (1). In the same formula, S + a = A I/ (A I + A *), S
IA=A */ (A t+ A t) ~ S (+1
'-, 1B=A (n-N/(A jn-11+ An
), S+n-na=An/1n-u+An. Furthermore, the memory item indicates an item that stores the calculation result one step before the current value. To calculate such a relational expression, give an initial value O to the memory item, apply voltage E to each memory item, calculate it sequentially at the fundamental period up to the lines T and -Tn, and calculate the final stage current 1 nB at the basic period. , articulatory voice output data can be obtained at this current of 1 nB.

Here, the equivalent circuit in Figure 2 (2) becomes a linear circuit, and the superposition principle holds true, and the current at each point when the voltage E is set to zero and the voltage EL is applied in series to the radiation impedance Zt can be found. In other words, the current of impedance Eo can be determined. In this way, by changing the surge impedance, the cross-sectional area of the vocal tract can be changed in a simulated manner, so by continuously changing the surge impedance, the articulatory coupling between phonemes or syllables is performed smoothly. .

Therefore, the present invention is characterized by interpolating the cross-sectional area changes of the acoustic tubes connected in series not only in the time axis direction but also in the vocal tract direction to simulate the movement of the tongue in the oral cavity.

F. Embodiment FIG. 1 is a diagram showing a block configuration of an embodiment of the present invention. 3 is a Japanese language processing unit, which converts the input Japanese text into pronunciations, etc. by referring to a dictionary for segmentation of phrases. Reference numeral 2 denotes a sentence processing section which performs processing for adding intonation to sentences. 3 is a syllable processing unit, which adds accents to the syllables that make up a sentence according to the intonation. 4 is a phoneme processing unit; 4. is a syllable parameter storage unit, and the phoneme processing unit 4 refers to the data in the syllable parameter storage unit 4I that defines the correspondence between syllables and phonemes, which are units of vowels and consonants, for input syllable data. into phonemes.

5 is a parameter interpolation processing unit; 5. is a phoneme parameter storage unit; is a sound source parameter storage section.

As shown in FIG. 3, the phoneme parameter storage unit 51 stores a plurality of utterance times of each phoneme, for example, the rising part t! .

It is divided into three time periods: steady its, and falling part tg, and for each time period, the duration, the pitch which is the repetition frequency of the sound source wave, the initial values of the energy of this sound source wave, the cross-sectional area of the sound tube, and the relevant time are determined. It stores time constants and sound source wave patterns that define how each initial value of a band changes to each initial value of the next time period. In this example, the human vocal tract (
The model is made by connecting 17 acoustic tubes with a length of 1 cm, and therefore, the cross-sectional area value is determined to be 17 pieces (Al to AI?) per one time period. The parameter supplementary law processing unit 5 processes each time period (t□
This is a part that performs interpolation processing of pitch, energy, and cross-sectional area in ~tm).

Reference numeral 6 denotes a calculation unit, which calculates the current i shown in FIG. 2 at the same timing as the interpolation calculation, for example, at a time interval of 100 μs, based on the parameters calculated by the parameter interpolation processing unit 5. Find the digital value of. 7 is a digital/analog (D/A) converter, which generates a current wave (analog current) based on the digital value obtained by the calculation section 6. Reference numeral 8 denotes a voice generating section such as a speaker, which generates voice based on analog current.

Next, specific interpolation calculations by the parameter interpolation processing section 5 will be explained.

For example, for a vowel, a rising part tx, a steady part t, 4. Falling part tg
, and the cross-sectional area parameter At(i
is the number of acoustic tubes (1-n), then 11. of the phoneme P.
1, configure the data for tg as shown in FIG. Similar data is configured for other phonemes Q as well.

Now, the cross-sectional area of these is the A1 phoneme and the P1 rising part! ,
When the order of the steady part M1 falling part E1 acoustic tube is 0 to 17, Ap□ for the phoneme P. , Ap□, Ap□...
・API+71 APN. ...'APMl? IApt
Similarly, for phoneme Q, o=Apg+ is Agxo-Aottt, AQMO-AQMI? ,
It is expressed as AQEO~Aog+7.

The cross-sectional area parameters of the acoustic tubes A0 to A17 are determined by the time parameters Tp and To for each phoneme P and Q from the phoneme processing unit in the previous stage.
is given. These time parameters Tp and To are determined by predetermined ratios for each phoneme (for example, rising part 1 of phoneme P = 1. steady part M = 3°, falling part E = 2. 1 of phoneme Q = 1. M = 2゜E=1) and the time is tx, t
M, allocate to tg.

When the phonemes P and Q are uttered in order from time 0 ■Times 0 to 1
Up to /6TP, the interpolation period is A2 every time.

The interpolation formula that changes the current value A1 with (the immediate cross-sectional area of the phoneme P in the sound tube i) as the target value is He5(t) = (Apwi-An(t-0)) ■ Time t
/6 TP to 4/6 T p, Apxi is set as the target value and the current value A is set as At(t)=at every interpolation period.
(Apwi-At(t=1/6Tp))■Time 4/
Until 6 T p= T p, the current value AI is set to Ap old as the target value every time At(t)= (APEI At(t=4/6Tp)
■ Until time T pA - T p + 1 / 4 T ta, the current value AI is set to AQl every t as the target value and AI (t) = (Aoy*-At (t = 'r
p))■Time Tp+1/4 Tg-Tp+3
/ 4 Until To, At(t) = [AQMI At
(t=Tp+1/4TQ))■Tp+3/4To-T
A1(t)=(Ao*t At(t
=Tp+3/4Tq)) The above is an example of linear interpolation, but
In addition to this, exponential approximation interpolation and the like can be similarly expressed.

These cross-sectional area parameters are calculated by assuming that n stages (17 stages in the example) of sound tubes divided at regular intervals in the vocal tract direction are continuous, and the cross-sectional area of each stage of sound tube is maintained for each phoneme. are doing. Then, the cross-sectional area values of each stage are smoothly interpolated in the time axis direction to generate a synthesized sound. Therefore, although the so-called mouth shape can be simulated well, the movement of the tongue within the oral cavity cannot be simulated well.

Therefore, the present invention further performs interpolation not only in the time axis direction but also in the vocal tract direction.

To simulate the movement of the tongue, interpolation is performed only within the range of the number of acoustic tube stages corresponding to the positions of the vocal tract where the tongue can move.

That is, assuming that the vocal tract position where the tongue can move is KI to KE, the tongue movement is interpolated within the range of KI to KE. In addition, here, KI to KE is 0≦Kl≦KE≦n-1(n
is the number of acoustic tube stages), and A at any time t
For a value of 1, when 0≦j≦KI and KE≦i≦n-1, the above-mentioned cross-sectional area interpolation is performed.

Next, this specific interpolation calculation will be explained.

The following calculation is performed within the range of KI≦i≦KE.

The cross-sectional area parameter A is set to the previous cross-sectional area parameter AA-1, with other conditions (phoneme 0 time) constant.

Compare with At+1. At this time, find the value of i (marked by X in FIG. 4) such that Ai<Ax 1 and A h <A h + 1, that is, A is the valley.

The value of this 1 is t PI t PM+ t Pg%E
It is taken as a time series in comparison with the phoneme data from the phoneme processing unit, such as Qll I QN#tQ11.

In this time series of i, the times before and after the change in i when the absolute value of the change Δi in i changes by 2 or more are defined as tα and tβ.

The time of tα and tβ is divided into 1Δ11 equal parts. Let the time be j+-Et...t1Δti-1t.

The cross-sectional area parameter AI of the valley at α and the valley A at tβ, +
Between Δt (dotted line section), the cross-sectional area parameters A++l, A
The cross-sectional area parameters of t+2...A1+Δ1-1 are interpolated in the same manner as in the above-mentioned normal method up to time tα.

Between time tα and tβ, interpolation is performed as follows.

■When AI+1, from tα to (tα+11), A, +1 (1) ・ (A, +l (t I)
-A1+1 (Jin α) )1-1 α × +Al+1 (t=tα) 1, from -1α (tα+1+) to tβ, As+1 (t) □ (At+j(tβ) -At
+1 (t a)) ■When A, +2, from tα to (tα+11), At+2(t)=(At+ 2(t t) At+
1 (tα)) 1-1 α X +A I+ 2 (L = tα
)11 1 α (tα+t,) to tβ is At+ 2(t) = (At+ 2(t β)
At+I(t α) and the following calculations are performed in the same manner.

That is, interpolation calculations are performed not only in the time axis direction but also in the vocal tract direction.

The above interpolation calculations can be performed without changing any phoneme parameters.

In addition, when articulating between the two sounds mp and Q, the vocal tract position Kl-K
If the minimum value positions of each Ai between E are close, the movement of the tongue will be small, and conversely, if the positions are far apart, the movement will be large.

Effects of the G1 Invention As described above, since interpolation calculations can be performed not only in the time axis direction but also in the vocal tract direction, tongue movement can be simulated, and as a result, articulatory combination becomes smoother and natural synthesized speech can be generated. It is particularly effective in the articulation of vowels.

Furthermore, since it can be implemented without changing the phoneme parameters in any way, the memory capacity for storing phoneme parameters can be as small as before.

In addition, when articulating between phonemes, the closer the minimum position of each A between the vocal tract positions Kl to KE, the less tongue movement there will be, and the farther away the tongue movement will be, so the utterance time length should be adjusted accordingly. It can make it look more natural.

[Brief explanation of the drawing]

Fig. 1 is an equivalent circuit diagram of an embodiment of the present invention, Fig. 2 is an equivalent model of an acoustic tube and a diagram of calculation modes, Fig. 3 is an explanatory diagram of phoneme parameters, and Fig. 4 is an explanation of the present invention. A diagram showing changes in the vocal tract cross-sectional area in the oral cavity is shown. 1... Japanese language processing section, 2... Text processing section, 3...
Syllable processing unit, 4... Phoneme processing unit, 5... Parameter supplementary processing unit, 6... Calculation unit, 7... D/A conversion unit, 8
...Voice part. Fig. 1 Equivalent circuit diagram of the embodiment Fig. 4 tα tβ

Claims (1)

[Claims] (1) By regarding the human vocal tract as a plurality of acoustic tubes connected in tandem, and by associating the acoustic tube group with the circuit element group of the surge impedance component and making the sound source correspond with the current source, the acoustic tube In a speech synthesis method that simulates speech waves emitted from the output end of a group of circuit elements based on power waves at the output end of a group of circuit elements, the utterance time of each phoneme is divided into one or more time periods for each phoneme that makes up a syllable. For each time period, the initial values of the pitch, which is the repetition frequency of the sound source wave, the energy of this sound source wave, and the cross-sectional area of the sound tube, and the initial values X_0 of the relevant time period, and the values of the next time period. a phoneme parameter storage unit that stores constants and sound source wave patterns that define how to change to each initial value X_r; and the pitch corresponding to the input phoneme data;
a parameter interpolation processing section that selects each initial value of energy and cross-sectional area from the phoneme parameter storage section and performs interpolation processing on these initial values; a calculation unit that calculates a current value output from the output end of the circuit element group based on the sound source wave pattern in the phoneme parameter storage unit;
and a vocalization section that generates a sound based on the calculation result of the calculation section, and the parameter interpolation section is configured to perform each front and rear cut-off within the range of the number of stages of the sound tube corresponding to the vocal tract position where the tongue can move. Compare the area parameters, take out the value of the cross-sectional area that is smaller than both the previous stage and the subsequent stage and form a valley as a time series of the sounds generated by consecutive phonemes, and then select an arbitrary value before and after when the absolute value of the change changes by 2 or more. A speech synthesis method characterized by dividing time equally, and interpolating the cross-sectional area by joining the value of the change for each equally divided time.