JPS6038720B2 - speech synthesizer - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a speech synthesis device that extracts a plurality of parameters characterizing speech from speech information and synthesizes speech from the parameters.

An example of a conventional PARCOR type speech synthesizer will be explained with reference to FIG.

The audio data ROMI, which stores multiple feature parameters of the audio signal analyzed and extracted from the original audio signal at fixed time intervals (hereinafter referred to as frame periods), and the microcomputer 2, which controls the operation of the system, process the audio data. It is connected to the synthesis circuit 3. When synthesizing voices, first, a signal is input from the microcomputer 2 to the microcomputer interface 4 for designating the address of the ROMI where the data of the voice to be uttered is stored. The microcomputer interface 4 then outputs a signal corresponding to the above signal to the ROM interface 5 to drive the ROM interface 5. In response to this signal, the ROM interface 6 extracts parameters from the audio data ROMI for each frame period. This parameter is determined by the next judgment circuit 6.
It determines whether there is a voice or not, whether parameter values are repeated, etc., and outputs the necessary parameters. Next, the conversion ROM 7 decodes these parameters. Furthermore, the interpolation circuit 8 smoothes the discrete changes in this parameter for each frame period. For example, linear interpolation is performed at time intervals shorter than the frame period. These parameters sequentially control the sound source 9 and the digital filter 10 to restore and synthesize the original sound, and the sound is generated from the external speaker 12 via the DA converter 11. The interpolation method of the interpolation circuit of the conventional speech synthesizer configured as above will be explained.

The Okema circuit 8 receives the daily frame parameters and new frame parameters extracted from the audio signal at one frame intervals. FIG. 2 shows how interpolation is performed at one frame intervals. The first value, which is the value of the old frame parameter, is set to A
, the linear darkness values CI to CN are calculated at regular time intervals toward the target value B, which is the value of the new frame parameter.
-, is output as the parameter value. When the target value B is reached after one frame time has elapsed, the next cycle starts and this new frame parameter is input as the old frame parameter and becomes the first value, and the new frame parameter of the next frame is used as the target value and the same process is performed. Interpolation will be performed.

In this interpolation method, as shown in Figure 3, the silent part 3
The last frame parameter 31 of 3 and the sound part 34
Interval is also performed between the parameter 32 of the first frame of

However, by such interpolation, the restored audio is
It has been confirmed through experiments that this gives an unnatural impression. Originally, at the start of speech, it is thought that the phonation mechanism changes discontinuously from a silent state to a voiced state, and it is considered that parameter values should also be interpolated from certain discontinuous values. SUMMARY OF THE INVENTION An object of the present invention is to provide a speech synthesizer having an interpolation method that can improve the disadvantages of synthesized speech by conventional speech synthesizers and synthesize more natural speech.

The speech synthesis device of the present invention is controlled by a leading value data storage section in which data of leading values of predetermined interpolation parameters are stored, and a signal for determining speech silence/speech, An interpolation input control circuit is provided in the preceding stage of the interpolation circuit, which inputs the data to the interpolation circuit as necessary.

With such a configuration, in the speech synthesis device of the present invention, when the original speech signal changes from silence to speech in the parameter interpolation method between frame periods, the signal sent from the determination circuit is used for interpolation input control. Controls the circuit, switches the old parameters input to the interpolation circuit, and the old parameter values of the new parameters to the predetermined parameters stored in the first value data section and inputs them to the interpolation circuit. and performs interpolation.

As a result, when the frame parameters move from a silent part to a sound part, the interpolation of the parameters in the time between them is performed using a predetermined parameter value as the leading value, rather than the frame parameter value of the silent part. As a result, the parameter value changes discontinuously to a specific value at the beginning of the sound. According to the speech synthesis device of the present invention as described above, compared to the conventional one,
It has been experimentally confirmed that natural speech can be synthesized, and the sound quality can be improved.

Furthermore, the configuration is simple and only requires adding a simple configuration to a conventional speech synthesis device. Next, one embodiment of the present invention will be described with reference to the drawings.

FIG. 4 is a block diagram showing an embodiment of the speech synthesis device of the present invention. Audio data-ROM 41 that stores parameter values for each frame representing the characteristics of the signal extracted by analyzing the original audio signal for each frame period in advance.
A microcomputer 42 that controls the operation of the entire system is connected to a speech synthesis circuit 43. The parameters for voice synthesis using the PARCOR method are normally 12: amplitude information Am, pitch information P, and PARCOR coefficients k, to k, and o. The mid-range information Am and the pitch information P provide information regarding the mid-range and pitch of the original signal that is the sound source of the synthesized speech.
Furthermore, the PARCOR coefficient physically means the reflection coefficient of the vocal tract, which controls the digital filter to electrically cause the sound source signal to reflect and resonate, thereby simulating the function of the actual vocal tract. Generally, the mid-swing information Am is a coefficient of o~, and the pitch information P is also a coefficient of o~. Also,
The PARCOR coefficients k, ~k, and o are each represented by a coefficient of -1 to 11.

The frame period is 10 seconds, and the parameters are encoded and stored in 48 bits per frame with 4 bits each. Furthermore, in the stationary audio portion, the parameters have almost the same values, so there is no need to provide repeating bits to store the same data. The storage capacity of the audio data ROM 41 is class 12 bits. This allows you to memorize approximately 59 seconds of audio. In the speech synthesis circuit 43, the microcomputer 42 specifies the starting address of the speech data to be produced, and the operating conditions of the speech synthesis chip, such as the start and stop frame period of speech, and the number of bits allocated per frame. A setting command is sent to the microcomputer interface 44. Microcomputer interface 4
4, a corresponding signal is sent to the ROM interface 45. The ROM interface 45 takes in necessary parameters from the audio data ROM 41 in response to the signal from the microcomputer interface 44, and the determination circuit 46 determines whether the sound is silent or harmful, and whether the sound is in a steady state. These parameters transform R
The data is sequentially sent to the OM 47, and in the case of 48 bits/frame, it is decoded into 10 bits each to facilitate subsequent processing. These data are sent to the interpolation input control circuit 48. The interpolation input control circuit 48 is controlled by the determination signal from the determination circuit 46 as to whether there is a sound or not, and converts the parameter input to the interpolation circuit 51 into a ROM.
It has a function of switching between the 47 side and the ROM 50 side where specific data is stored. Normally, the parameters from the conversion ROM 47 are sent to the interpolation circuit 51 as they are.
However, when the original audio signal changes from silence to sound,
The signal from the determination circuit 46 is transmitted to this interpolation input control circuit 48.
The data is sent to the head value data ROM control circuit 49, which controls it. The ROM control circuit 49 takes in predetermined parameter values stored in the ROM 501 in response to the signal from the determination circuit 46 and sends them to the interpolation input control circuit 48 . According to our experiments, “As a predetermined parameter, the amplitude Am=OPARCO
The R coefficient is k,=0.97, k2=-0.80k3=0
．． 40, k4; -0.20, k5=0.06, k6=-
0.13, ks=0.18, and k,o=-0.08 were found to be the best.

The interpolation input control circuit 48 sends these parameters to the Kusuma circuit 51. The interpolation circuit 51 linearly interpolates the parameters inputted from the interpolation circuit input control circuit 48 at a frame interval of 10 ms every 2.5 seconds.
This smoothes out the discrete changes in the feature parameters from frame to frame. The pitch information P controls the sound source section 52 to generate a sound source signal, and the amplitude information Am and the PARCOR coefficients k, to k, o control the digital filter 53 to generate a sound source signal from the sound source section 52 to a predetermined value. Convert to audio. The digital filter 53 has a structure in which 10 stages of lattice filter circuits are connected in cascade, and a PARCOR coefficient is used as a filter coefficient of each lattice filter circuit. As a result, the original voice is restored and synthesized, and the D-A converter 54
Sound is generated from an external speaker 55 via the external speaker 55. Next, the structure and function of the interpolation input control circuit and the interpolation circuit used in one embodiment of the present invention are shown in FIG. 6 and FIGS. 6a to 6b.
Explain using.

The interpolation input control circuit 48 receives parameter values at one frame interval as input. Old frame parameter input terminal 5-6 and new frame parameter input terminal 57 and RO
There is a specific value input terminal 58 into which a predetermined specific value from M50 is input. The signal from the old frame parameter input terminal 56 is controlled by a switch 59,
The signal from the specific value input terminal 58 is sent to another switch 6
Controlled by 0. These switches are, for example, MOS
Consists of FET. FIGS. 6a and 6b are timing charts of signals sent from the determination circuit to control these switches 59 and 60. A signal 66 controlling the gate of the old frame parameter input switch 59 normally has an output of 1, but in a frame section 68 where the original audio signal changes from silence to sound, the output becomes 0 and the switch 59 is opened. Conversely, a signal 67 that controls the specific value input switch 60
normally has an output of 0, but in that section 68, the output is 1.
, and close the switch 60. The interpolation circuit 51 normally inputs the old parameter value as the leading value and the new parameter value as the target value to the subtraction circuit 63, but in a frame in which the original audio signal changes from silence to sound, the above-mentioned switch 59. By the function 60, specific data is input as the leading value. A subtraction circuit 63 calculates these differences, and then a division circuit 64 divides this difference by, for example, 1/8 to calculate a parameter increase or decrease value of 2.5 ms for one frame of 20 ms, and an adder circuit 65 is added to the leading value to output the first darkness correction value. Next, switch the switch 66 to feed back the world power value,
The parameter increase value is added again by the addition circuit 65 and output. In this way, interpolation is performed up to the target value. An example of parameter interpolation using the interpolation method of the present invention as described above will be explained with reference to FIG.

By the interpolation input control circuit, in order to configure the start part of the voice when changing from the non-capital part 71 to the sound part 72,
A specific value 74 is input as the old parameter. Using that value as the leading value, a new parameter value of 75 is used as the target value. Therefore, in this frame, the parameter value changes to a discontinuous value and is interpolated from the silent part to the sound part, instead of the smooth interval between the buckets as in the conventional case. It has been experimentally confirmed that the speech synthesis device of the present invention as described above can synthesize more natural speech than conventional devices.

Moreover, the structure is very simple and can be easily added to conventional ones.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing a conventional speech synthesis device, FIGS. 2 and 3 are explanatory diagrams showing an interpolation method of the Tachibana circuit of the conventional speech synthesis device, and FIG. 4 is a block diagram showing a conventional speech synthesis device. 5 is a block diagram showing a speech synthesizer according to the present invention, FIG. 5 is a circuit diagram showing a part of the speech synthesizer according to the present invention, and FIG. 6 is a block diagram showing signals for controlling the circuit shown in FIG. FIG. 7 is an explanatory diagram for explaining the interpolation method of the speech synthesizer according to the present invention. 41...Audio data-ROM, 45...
・ROM interface, 46...determination circuit, 47...conversion ROM, 48...interpolation input control circuit, 49...ROM control circuit, 50...
...ROM, 51...Interpolation circuit, 52...
...Sound source, 53...Digital filter, 54. ．． …． D/A converter, 55...Speech power. Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7

Claims (1)

[Claims] 1. An input section into which a plurality of parameters representing voice characteristics are input at fixed time intervals, an interpolation circuit connected to the input section and interpolating parameter values during the fixed time cycles, and a predetermined a data storage unit in which specific parameters are stored; an interpolation input control circuit connected to the data storage unit and configured to send the specific parameters to the interpolation circuit when the audio changes from silent to active; 1. A speech synthesis device comprising: a circuit that generates a speech signal based on parameters interpolated by an interpolation circuit.