JP2614436B2 - Speech synthesizer - Google Patents

Description

The present invention relates to a speech synthesizer for synthesizing speech using a parameter representing characteristics of a sound source and a vocal tract.

B) Conventional technology In speech synthesis, sound source waveforms are compressed by performing differences, adaptive quantization, and the like, compressed and stored as speech data, and waveform processing methods such as Δ-PCM and ADPCM used for speech synthesis are used. As shown in Japanese Patent Publication No. 52-7282, there is a generator type such as LPC, PARCOR, etc., which extracts a parameter representing the spectral envelope of the vocal tract and synthesizes a voice from the parameter and a sound source wave. At present, a source-type method having a high data compression rate is predominant.

In a general method such as LPC and PARCOR, voice is analyzed to extract data such as a coefficient such as LPC and PARCOR, a fundamental frequency (pitch), and an amplitude at regular intervals, and store the data as voice data in a memory. This fixed time is called a frame period. At the time of synthesis, the sound source wave (1) is generated from the fundamental frequency (pitch) data and the amplitude stored as audio data, as schematically shown in FIG. Digital filter (3) composed of this sound source wave (1) by coefficients (2) such as LPC and PARCOR
, A speech waveform (4) is synthesized.

Conventionally, as this sound source wave, an impulse wave, a triangular wave,
A residual wave is used. Of these, the use of residual waves allows the original sound to be reproduced more faithfully and provides better sound quality than the case of using impulse waves and triangular waves, but requires a large amount of memory to store the residual waves. There was a disadvantage that required.

Various attempts have been made at present to reduce the memory required to store the residual wave, but there is still a limit to the reduction in the required memory.

(C) Problems to be Solved by the Invention The present invention has been made in view of the above points, and basically uses an impulse wave, a triangular wave, or the like as a sound source and reduces the amount of memory while increasing the fundamental frequency of the original sound. They are faithfully reproduced.

Here, the problem will be described using an example in which an impulse wave is used as a sound source.

FIG. 3 illustrates a conventional speech synthesis process, in which an impulse wave (5) is used as a sound source when the original voice is a voiced portion, and a white noise (6) is used when the original voice is a voiceless portion. (7) is a switch for switching the sound source between voiced sound and unvoiced sound.

FIG. 4 shows a sound source waveform in a voiced sound portion. In FIG. 4, the horizontal axis represents time t, and the vertical axis represents amplitude A. Then, the interval P between the impulse waves becomes the fundamental frequency.

However, in the conventional method, since the impulse wave is generated at a timing synchronized with the sampling frequency (for example, 10 KHz), the interval (pitch) between the impulse waves is
It is an integral multiple of 100 μsec, and can only take the value of the fundamental frequency of 10,000 / n Hz (n is an integer). Therefore, even if the original voice having the fundamental frequency pattern as shown in FIG. 5 is analyzed and the fundamental frequency is correctly extracted, the synthesized voice has only a low-accuracy fundamental frequency pattern as shown in FIG. Could not be obtained.

D) Means for Solving the Problems The speech synthesizer of the present invention extracts parameters representing the fundamental frequency of speech and the characteristics of the vocal tract from the original speech in advance, stores them in memory, and reads them from the memory. In a speech synthesizer for synthesizing speech by generating a sound source wave from a fundamental frequency and applying the sound source wave to a digital filter having a parameter representing a characteristic of a vocal tract as a coefficient, the sound source wave is a reciprocal of the fundamental frequency. The basic waveform has a basic period, and the basic waveform is composed of two pulse waves adjacent to a predetermined position. In each of the basic waveforms,
By controlling the amplitudes of the two adjacent pulse waves independently variable, the center of gravity of the two adjacent pulse waves is changed, and the frequency value is changed within a predetermined frequency centered on the fundamental frequency. It is characterized by speech synthesis at least.

E) Operation According to the speech synthesizer of the present invention, the sound source wave is composed of a basic waveform having a reciprocal of the fundamental frequency in a fundamental period, and the basic waveform is composed of two pulse waves adjacent to a predetermined position. In the waveform, by controlling the amplitude of two adjacent pulse waves independently variable, the center of gravity of the two adjacent pulse waves is changed, and the frequency value fluctuates within a predetermined frequency centered on the fundamental frequency. At the very least, we will synthesize speech,
A more natural synthesis of human voice becomes possible.

F) Embodiment As shown in FIG. 7, for example, a sound source wave used in the speech synthesizer of the present invention has two continuous impulse waves (a),
(B), and the ratio α (= Ab / A) of the amplitude Aa of the first impulse wave (a) to the amplitude Ab of the second impulse wave (b)
a) is represented by, for example, 2-bit data. In this case, the sound source wave when the value of α changes to 0, 1/3, 1, 3 is
As shown in FIGS. 8 (i), (ii), (iii) and (iv), the center of gravity indicated by these arrows moves between both impulse waves (a) and (b). In FIG. 8, the horizontal axis represents time, and the vertical axis represents waveform amplitude.

In FIG. 8, assuming that the sampling frequency is 8 kHz,
If 40 samples are taken, the basic period is T = 5 msec (basic frequency: F = 200 Hz). Here, assuming that the waveforms as shown in FIG. 8 are continuous, the center of gravity of each waveform is at the position of the arrow, and the fundamental frequency is the distance of the center of gravity between the waveforms.
That is, it is the reciprocal of the distance between arrows (basic period). Specifically, F1 = 1 / T1, F2 = 1 / T2, and F3 = 1 / T3 are the fundamental frequencies.

Therefore, T1 <T, T2> T, and T3> T, and these are converted into frequencies. If the vertical axis is frequency and the horizontal axis is time, the result is as shown in FIG. That is, the frequencies F1, F2, and F3 are arranged in a state where they are varied within a predetermined frequency centered on a certain fundamental frequency F.
A more natural synthesis of human voice becomes possible.

FIG. 1 shows an embodiment of the voice synthesizing apparatus of the present invention in which the above-mentioned sound source wave control is performed. Next, the operation of this apparatus will be described.

When a voice output instruction is issued from the outside to the memory read control unit (9), the memory read control unit (9) stores the head address of the voice data stored in the voice data memory (11) in the address counter (10). The contents of the audio data memory (11) are sequentially transferred to the buffer (12).

The content of the audio data stored in the audio data memory (11) and the buffer (12) is as shown in FIG. 10, where α is the amplitude ratio of the adjacent impulse waves described above,
By changing the center of gravity of the impulse wave, the fundamental frequency can be more finely controlled. In the figure, P indicates fundamental frequency data (pitch interval), A indicates amplitude data, and K0 to K9 indicate PARCOR coefficients. Buffer (12)
Are transferred to each of the parameters α, P, A,
It is divided into K0 to K9, and A
Register (13), α register (14), P register (15)
Each is stored.

On the other hand, the PARCOR coefficients K0 to K9 are digital filters (16)
And used as coefficients. In the impulse wave generator (17), the A register (13), the α register (14),
According to the contents of the P register (15), the eighth
An impulse wave of any one of the forms (i) to (iv) is generated.

According to the contents of the P register (15), the sound source switching unit (18) determines whether or not the sound is a voiced sound by an impulse wave generation unit (17).
If the output is unvoiced, the sound source is switched so that the output of the white noise generator (19) is given as an input to the digital filter (16). The output of the digital filter (16) is output as a synthesized voice through the DA converter (20).

G) Effects of the Invention As is clear from the above description, according to the speech synthesis apparatus of the present invention, the parameters representing the fundamental frequency of the speech and the characteristics of the vocal tract are extracted from the original speech in advance and stored in the memory. A sound synthesizer for synthesizing a sound by generating a sound source wave from a fundamental frequency read from a memory and applying the sound source wave to a digital filter having coefficients representing parameters of vocal tract characteristics as coefficients. Is composed of a basic waveform having a reciprocal of the fundamental frequency in a fundamental cycle, and the basic waveform is composed of two pulse waves adjacent to a predetermined position.
By controlling the amplitude of each pulse wave to be independently variable,
Since the voice synthesis is performed by changing the center of gravity of the two adjacent pulse waves and changing the frequency value within a predetermined frequency centering on the fundamental frequency, more natural human voice synthesis is possible. .

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a speech synthesizer of the present invention, FIGS. 2 and 3 are schematic diagrams showing a speech synthesis method, respectively, FIG. 4 is an impulse waveform diagram used as a sound source,
FIG. 5 is a diagram of a fundamental frequency characteristic of an original voice, FIG. 6 is a diagram of a fundamental frequency pattern of a voice synthesized by a conventional method, FIG. 7 is a diagram of a sound source waveform used in the present invention, and FIG. ) (Ii
i) and (iv) are sound source waveform diagrams each showing a change in the sound source waveform diagram of FIG. 7, FIG. 9 is a fundamental frequency pattern diagram of speech synthesized by the speech synthesizer of the present invention, and FIG. 10 is a memory diagram. is there. (11) ... voice data memory, (13) ... A register,
(14)… α register, (15)… P register, (16)
... A digital filter, (17) an impulse wave generator, (19) a white noise generator.

Claims (1)

(57) [Claims] 1. A fundamental frequency of a voice and parameters representing characteristics of a vocal tract are extracted in advance from an original voice and stored in a memory, and a sound source wave is generated from the fundamental frequency read from the memory. Is applied to a digital filter that uses a parameter representing a characteristic of a vocal tract as a coefficient, thereby synthesizing a voice. In the speech synthesizer, the sound source wave includes a fundamental waveform having a reciprocal of the fundamental frequency in a fundamental cycle; Is composed of two pulse waves adjacent to a predetermined position. In each of the basic waveforms, the amplitude of the two adjacent pulse waves is independently variably controlled, so that the center of gravity of the two adjacent pulse waves is A speech synthesizer characterized by changing a point and changing a frequency value within a predetermined frequency centered on the fundamental frequency to synthesize speech.