JP2572961B2 - Voice synthesis method - Google Patents

Description

The present invention relates to a speech synthesizer. In particular, it relates to a circuit for generating an unvoiced sound source.

(B) Conventional technology Various speech synthesis methods have been proposed as shown in the article "Comparison of Various Speech Synthesis Methods Familiar with You" published by Nikkei Electronics on February 4, 1980. In general, a waveform processing method that compresses information in the form of a voice waveform as it is, and a parameter representing a fundamental period (referred to as a pitch) of a sound source and a transfer characteristic of a vocal tract are extracted.
It is roughly classified into a sound source and a generation source method for synthesizing speech by the parameter. In recent years, the generation source method is becoming mainstream because of its high information compression ratio and good quality of reproduced sound. FIG. 3 shows an outline of a speech synthesizer of a generation source system. In the figure, (1) is a voiced sound source generation unit, which generates a sound source signal P corresponding to the pitch amplitude. (2)
The unvoiced sound source generation unit generates a sound source signal N corresponding to an expiratory flow that can be regarded as white noise when the vocal cords are not vibrated. (3) is a switch for switching the sound source signals P and N from the two sound source generation units (1) and (2) according to the voiced / unvoiced judgment information D. (4) is a digital filter unit corresponding to the transfer characteristic of the vocal tract, which generates and outputs a synthesized voice V in response to the input of the sound source signals N and P from the switch (3).

As the unvoiced sound generator (2) in such a speech synthesizer, a conventional M sequence (maximum length null sequence)
, And this counter usually has a configuration of 10 bits or more.

(C) Problems to be Solved by the Invention In a conventional speech synthesis method employing an unvoiced sound generation section having an M-sequence feedback counter configuration, the M-sequence feedback counter in the unvoiced sound generation section has a number of constituent bits. If the number is increased, the output cycle can be lengthened, that is, the time length of the noise period within one cycle can be increased, but white noise information cannot be completely obtained. Therefore, when synthesizing an unvoiced sound having a large continuous length, periodicity appears and the quality of synthesized speech is deteriorated.

(D) Means for Solving the Problems The speech synthesis method of the present invention uses the conventional feedback counter as the unvoiced sound source generation unit, The noise information of the unvoiced sound is obtained by the arithmetic processing of the recurrence formula.

(E) Action The above recurrence formula is a simple fixed-difference equation caused by May-proposed Figungpaum bifurcation (detailed in Mathematical Science No. 207, SEPTEMBER, 1980). You can get information.

(F) Embodiment FIG. 1 shows a functional configuration of an unvoiced sound source generator used in the speech synthesis method of the present invention. In the figure, (5) is (1)
−xn), (6) is a multiplier for calculating axn, (7) is a multiplier for calculating axn (1−xn),
This value xn + i = axn (1-xn) is temporarily stored in the register (8) and then fed back at the next timing (n + 1) to be input to the subtracter (5).

Thus, the unvoiced sound source generation unit in FIG. 1 has xn + i = axn (1
−xn) is sequentially calculated to generate a sequence {x _n }, and this sequence {x _n } is used as amplitude information of noise information.
The initial value x0 for the operation of this sequence {x _n } is defined as o <x0 <1
It is necessary to The specific value of the constant a for making the sequence {xn} white noise is 3.9a <4
It has been experimentally found that the closer to 4 in the range, the better. FIG. 2 shows that a = 3.99 and n
= 0,1,2, ..., 255 shows the frequency characteristics when the Fourier transform is performed on 256 sample strings {xn}. To compare with this, Fig. 4 shows the conventional 10-bit under the same conditions. FIG. 4 shows a frequency characteristic diagram of an output sample sequence of an M-sequence feedback counter having a configuration.

Comparing these figures, it can be seen that the noise information of the unvoiced sound source obtained by the method of the present invention can be flattened and white noise with less periodicity than that of the conventional method.

In the above description of FIG. 1, the multipliers (6) and (7) are shown individually for the sake of explanation. However, a single multiplier can be used in a time-sharing manner, and the configuration is simple. The present invention can be implemented. Further, it is not necessary to provide a dedicated multiplier for the unvoiced sound source generating unit, and the multiplier provided in the digital filter unit can be time-divisionally used. For example, in the case of the Percall speech synthesis method, the multiplier in this filter unit normally performs about 20 multiplications in one sample period for synthesizing one sample of the speech signal. During one sample period, 2
You only need to add multiplications.

(G) Effect of the Invention As is clear from the above description, the speech synthesis method of the present invention has found that the equation proposed by May may be employed in an unvoiced sound source generation unit for speech synthesis. Accordingly, the unvoiced sound source signal can be a white noise signal having no periodicity, and the quality of unvoiced sound can be greatly improved, so that a more natural synthesized voice can be obtained. Moreover, if the method of the present invention is used, the configuration of the unvoiced sound source generation unit can be simplified, and as a result, the cost of the speech synthesizer can be reduced.

[Brief description of the drawings]

FIG. 1 is a functional configuration diagram of an unvoiced sound source generating unit when the voice synthesis method of the present invention is used, FIG. 2 is a frequency characteristic diagram of an unvoiced sound source signal when the method of the present invention is used, and FIG. FIG. 4 is a diagram showing a frequency characteristic of an unvoiced sound source signal according to a conventional method. (2)… unvoiced sound source generation unit, (5)… subtractor, (6)
(7)… Multiplier, (8)… Register.

Claims (1)

(57) [Claims] A sound source signal containing pitch information of a voiced sound from a voiced sound source generation unit and a sound source signal containing unvoiced sound noise information from an unvoiced sound source generation unit are input to a vocal tract characteristic filter unit. In the speech synthesis method for reproducing and outputting a speech signal by adding a vocal tract characteristic to the sound source signal, the unvoiced sound source generation unit includes: x _{n + 1} = ax _n (1-x _n ) n = 0,1 , 2,3,... 0 <x ₀ <1 a is a constant The operation of a recurrence formula is performed, and the sequence {x _n } obtained by the operation is used as amplitude information of the noise information. Speech synthesis method.