JPH0141999B2 - - Google Patents

Description

[Detailed description of the invention]

The present invention relates to a speech synthesis device. In general, the characteristic parameters that represent the characteristics of speech include the amplitude parameter (hereinafter referred to as the A parameter) that represents the magnitude of the sound, the pitch parameter (hereinafter referred to as the P parameter) that represents the pitch or fundamental period of the sound, and the pitch parameter that represents the pitch of the sound (hereinafter referred to as the P parameter). There are spectral parameters (hereinafter abbreviated as S-parameters) that represent the timbre or spectral distribution of . Therefore, in order to synthesize speech, the speech signal is sampled with a sampling pulse having a frequency sufficiently higher than the speech frequency, each feature parameter is extracted and stored in a data memory in advance, and the feature parameters read from the data memory are All you have to do is drive the sound source based on this and synthesize the sound. In this type of speech synthesis device, the greater the number of samplings of the audio signal, the more faithful the speech can be synthesized. However, on the other hand, as the number of samplings increases, the number of bits of the speech synthesis data increases, requiring a large capacity data memory. There is a problem in that the data processing circuit configuration becomes complicated and the cost increases as the data processing becomes necessary. Therefore, in the case of conventional speech synthesis devices, the sampling pulse frequency (hereinafter abbreviated as sampling frequency)
is set to the minimum frequency required to faithfully reproduce the human voice, and the sampling frequency is usually set to 8 or 10 KHz (sampling cycle 125 μS or 100 μS). By the way, by sampling the audio signal with the sampling pulse, A,
When extracting feature parameters consisting of P and S parameters and storing them in a memory, and reading out the feature parameters stored in the memory using a synchronization pulse with a period equal to the sampling pulse to synthesize speech,
The fundamental period of audio reproduced based on the P parameter can only take discrete values determined by the sampling frequency. In other words, the sampling period is
100μS, and the P parameter is Pi (an integer value), the fundamental period t of reproduction is t = 100Pi × 10 ^-6 (sec) (however, Pi = 1, 2, 3...), and the reproducible audio frequency is The result will be discrete values as shown in Table 1.

【table】

[Table] Even if only such discrete audio frequencies can be generated, the human voice can be reproduced with relative fidelity.
However, when reproducing a melody sound composed of scale frequencies, the scale frequencies of each scale (C, D, E, etc.) are often not included in the above discrete values as shown in Table 1, and the melody sound is If such discrete audio frequencies were used for reproduction, there was a problem in that melody sounds with a significantly shifted pitch would be reproduced. The present invention aims to solve the above problems. An embodiment of the PARCOR type speech synthesizer will be described below with reference to the drawings. As shown in Figure 1, the PARCOR type speech synthesis method samples the audio signal V _S at an appropriate period to using sampling pulses, and then converts the sampled value Xt to
PARCOR coefficient (partial autocorrelation coefficient: hereinafter abbreviated as K parameter) that extracts only the correlation between Xt and X _tp , excluding the correlation due to (P-1) sampling values between X _tp The sound is synthesized using the S parameter as the K parameter, and the K parameter is set at an appropriate period to (approximately
The audio signal V _S is sampled every 100 μsec), and the correlation coefficient between adjacent sample values is K ₁ .
Between multiple isolated sample values, the influence of the intervening sample values is determined by linear prediction using the least squares error, and the correlation coefficient obtained by subtracting them is K ₂ ~ _{K o} . be. This K
The parameters are K ₁ , K ₂ , and K ₃ , which represent partial autocorrelation with points close to X _t . Coefficients that represent partial autocorrelation with points close to X t contain a wealth of information about the spectral distribution, but K ₈ ,
_{Since the} partial autocorrelation coefficients with _points far from It is more effective to reduce the number of bits and reduce redundancy by allocating a small number of quantization bits to the K parameter. Therefore, the PARCOR method has a better band compression rate than the autocorrelation coefficient method, which uses autocorrelation coefficients as S-parameters and allocates the same number of bits to each coefficient.
Typically, each A, P, and K parameter is stored or transmitted in a compressed manner, with 5 bits for the A parameter, 6 bits for the P parameter, and 7 bits for each coefficient K ₁ , K _{2 .} . . K ₁₀ of the K parameter. ,6,5,
Assign 4, 4, 4, 3, 3, 3, 3 bits, etc. FIG. 2 is a block circuit diagram of an embodiment of a PARCOR type speech synthesizer used in time signal devices, alarm devices, alarm devices, etc., and shows a data memory M for storing voices and melodies as compressed feature parameters.
It consists of a control IC (A) equipped with a voice synthesis IC (dotted line areas A and B), and a voice synthesis IC (excluding the dotted lines A and B), and data is exchanged between the two ICs in a bit-serial manner. By the way, all voice characteristic parameters are stored as 10-bit data in the playback ROM 1, and the number of data allocated to each characteristic parameter is optimally distributed according to the degree to which that characteristic parameter contributes to sound quality. ing. For example, in the case of the A parameter, 32 pieces of data expressed in 10 bits are stored. Therefore, the number of relative address bits required when accessing arbitrary data of the A parameter is 5 bits. This relative address is called a compressed parameter because it represents the characteristic parameter compressed to the minimum necessary size. For playback
The actual feature parameters stored in ROM1 are called playback parameters. As is clear from the above, the number of bits of the playback parameter is A,
The characteristic parameters P, _K1 to _K10 all have 10 bits in common, but the number of bits of the compression parameter differs for each parameter A, P, _K1 to _K10 , for example 5 and 6, respectively. ,
3, 3, 3, 3, 4, 4, 4, 5, 6, 7 bits (53 bits in total). In addition, 3 bits, or 8 pieces of data, are reserved for playback.
It is secured in ROM1. Such a compression parameter is between 5 and 20, which allows the audio signal to be considered to be in an approximately steady state.
Since one set (=53 bits) is extracted every msec (one frame), it is possible to reproduce the audio signal by processing the data at a rate of at most 2650 bits/second, and even silent sections and repeat sections can be reproduced. Taking this into consideration, it is actually possible to reproduce audio signals at around 1600 bits/second. by the way,
In this example, we will discuss the fundamental period generation method for synthesizing speech that changes uniformly and continuously in pitch, such as spoken words, and for synthesizing speech that continues discretely, such as melody sounds and singing. When playing a melody sound, a melody control code is added to the beginning of the compression A parameter among the compression parameters input from the control IC (A) to the data input terminal 8, and the melody control code is Melody control code detection signal from the detection circuit 34
When V _M is obtained, the audio-melody switching circuit 33
is switched to the melody side (side b), and the sound source is driven at a basic period equal to the basic period of each scale note, thereby synthesizing melody sounds. The basic configuration and operation (normal speech synthesis operation for synthesizing human voice, etc.) of the embodiment will be described below. Now, the compression parameters (i.e. playback ROM
1 relative address) is stored bit-serially in the ring register 3 from the data input terminal 8 via the switching circuit 10 for each frame, but with only such a relative address, the playback data of each parameter is stored in the playback ROM 1. are stored consecutively, so specific data cannot be retrieved. Therefore, the start address of each parameter in the playback ROM stored in the index ROM2 is taken out one after another under the control of the address counter 11, and added to the above relative address by the adder circuit 4. An absolute address (9 bits) is calculated, and the playback ROM 1 is accessed using this absolute address. The index ROM 2 stores the bit allocation number of the compression parameter as a 3-bit binary number, and the data regarding the bit allocation number of the compression parameter is sent to the reproduction control circuit 12, and the reproduction control circuit 12 stores the bit allocation number of the compression parameter. Send the shift clock to ring register 3. Therefore, from the ring register 3, depending on the number of bit allocations mentioned above, for example, in the case of the A parameter, 5 bits, in the case of the P parameter, 6 bits, in the case of the _K10 parameter, ₃ bits,... If the compression parameter (relative address) is 7 bits,
are sent to the adder circuit serially. Furthermore, the starting address in the reproduction ROM 1 of each feature parameter stored in the index ROM 2 is determined by the parallel serial conversion circuit 1.
3, one bit at a time is sent to the adder circuit 4, so that the absolute address is calculated by sequentially adding one bit at a time. The serial absolute address calculated in this way is converted into parallel data via the serial-parallel conversion circuit 14,
It is converted into an address for accessing the playback ROM1. The feature parameters output from this playback ROM 1 are updated every frame, but if the feature parameters change discontinuously at the connection points between each frame when updating data, distortion may occur in the audio signal. Therefore, an interpolation calculation circuit 5 is provided to perform approximate linear interpolation at 8 points within one frame so that the feature parameters can change smoothly when updating data. I have to. Therefore, the timing control circuit 28 generates eight interpolation D clocks (2.5 msec) during one frame (20 msec), as shown in FIG.
P clock for reading parameters (100μsec), and 22 T clocks for reading bits in one P clock
Create a clock (4.5μsec cycle). Note that the P clock is a synchronous pulse corresponding to a sampling pulse. First D ₁ of 8 D clocks
from data input terminal 8 to ring register 3
The data is loaded into. Each compression parameter A,
P, _K10 ..., _K1 are read sequentially at odd-numbered P clocks. For example, the A parameter is
It is read using five T clocks from T ₆ to T ₁₀ in the P ₁ section. The even numbered P clocks or T clocks other than those mentioned above are used as timing for the interpolation calculation circuit 5, the sound source ROM 6, the digital filter 7, etc. This interpolation calculation circuit 5 stops its operation when a melody control code is detected. Each feature parameter updated to a new value every 2.5 msec by the interpolation calculation circuit 5 is temporarily stored in the P latch 16 and the AK latch 23, respectively. However, all parameters that are not needed for the time being are stored in the AK parameter stack 24.
The data is transferred to the digital filter 7 and stored as speech synthesis data. By the way, the P parameter stored in the P latch 16 is data for driving the voiced sound source 19 to generate an impulse signal having a fundamental period corresponding to the P parameter, and the output is obtained from the melody control code detection circuit 34. When the voice-melody switching circuit 33 is switched to the side (a side) that synthesizes voices such as human speech, the address counter 18 of the sound source ROM 6, which is counting the P clock equal to the sampling pulse, The reset signal is the output of a match circuit 17 that detects a match between the address counter 18 output and the P parameter stored in the P latch 16, and the address counter 18 is reset at a cycle that is an integral multiple of the P clock cycle (P parameter). It's getting old.
Therefore, the voiced sound source control data based on the P parameter is output from the sound source ROM 6, and the voiced sound source 19
An impulse signal having a fundamental period (discrete audio frequencies as shown in Table 1) corresponding to the P parameter is generated. Note that when the sound does not have a fundamental period, the sound source control circuit 20 switches the switching circuit 22
The unvoiced sound source 21 generates white noise having no fundamental period.
Next, the A parameter and the K parameter are supplied to the digital filter 7, which reproduces the voice by adding information about amplitude and spectral distribution to the signal supplied from the voiced and unvoiced sound sources. In the figure, 25 is a low frequency amplifier that amplifies the reproduced audio signal, 26 is a speaker, and 27
is a crystal oscillation circuit. Scale signal generation circuit 3 shown in FIGS. 4 to 6 below.
1. The configuration of the reset pulse generation circuit 32 and the voice synthesis operation for synthesizing melody sounds will be explained. The scale signal generation circuit 31 includes a shift register 31a that takes in data corresponding to the P parameter, that is, a compressed P parameter output from the control IC (A), using a request signal V _RE , and a shift register 31a that receives data corresponding to the P parameter, that is, a compressed P parameter output from the control IC (A), and a shift register 31a that receives the compressed P parameter as address data. A scale ROM 31b configured to read out scale data corresponding to the scale ROM 31b, and a preset counter 3 which receives the scale data read out from the scale ROM 31b as a preset input and counts clock pulses having a higher frequency than the P clock, such as the T clock.
1c and an inverter 31d that inverts the 0 detection signal of the preset counter 31c, and outputs a 0 detection signal having a cycle that is an integral multiple (scale data) of the clock pulse cycle as a scale signal P _M. In this case, the frequency of the scale signal P _M output from the scale signal generation circuit 31 takes discrete values, but the discrete interval becomes smaller in accordance with the frequency of the clock pulse. Therefore, by storing appropriate scale data in the scale ROM 31b, the scale signal generation circuit 31 can generate a scale signal P _M that matches the frequency of each scale signal. For example, the clock pulse is T clock (period: 4.5μsec).
If the scale data "284" is read from the scale ROM 31b using the compressed P parameter corresponding to the P parameter "12", a 0 detection signal is obtained from the preset counter 31c at a cycle of 4.5 x 284 μsec, and this The scale signal P _M has a fundamental period corresponding to "12.8" of the P parameter, and a discrete fundamental period corresponding to the P parameter can be interpolated. The reset pulse generating circuit 32 is formed by inverters 35a, 35b, a capacitor 36, a NAND gate 37, a D flip-flop 38, and a NAND gate 39, and as shown in the time chart of FIG.
The P clock immediately after obtaining the scale signal P _M output from the address counter 18 is output as the reset pulse V _R of the address counter 18. In the figure, A shows the output of the coincidence circuit 17 when the P parameter is "12", B shows the scale signal P _M , and C shows the reset pulse _VR . If the melody control code is now output from the control IC (A) and the melody control code detection signal V _M is obtained from the melody control code detection circuit 34, the audio-melody switching circuit 33 switches to the melody side (side b). The address counter 18 is reset by the reset pulse V _R output from the reset pulse generation circuit, and the address counter 18 is reset by counting 13 P clocks and reset by counting 12 P clocks. This will occur at a ratio of 4:1. Therefore, equivalently, the P parameter is “12.8”
The sound source ROM 6 is addressed at a fundamental period corresponding to , and the voiced sound source 19 is controlled, so that the scale note "G" is accurately reproduced. Similarly, each scale note is accurately reproduced, and the melody is reproduced at the correct pitch. The time chart shown in Figure 7b is the scale signal P _M
This explains the relationship between the reset pulse VR and the reset pulse V _R more clearly.
The reset pulse V _R corresponding to the scale signal P _M is shown. As is clear from the figure, P _pulses 3, 6, 8, 11,
The 14th, 16th... pulse is output. Since the sound source ROM 6 is addressed by the address counter 18 which is reset by this reset pulse _VR ,
Voiced sound source data is read from the sound source ROM 6 at a cycle that can equivalently be considered as 3.75KHz (800/3μsec), and the voiced sound source 19 is driven at the correct scale frequency, so that sounds such as melody sounds and singing are produced at accurate pitches. It will be played. In the embodiment, the compression parameters are used as the address data of the scale ROM 31b, but the P parameters stored in the P latch 16 may be used as the address of the scale ROM 31b. As described above, the present invention samples an audio signal with a sampling pulse having a frequency higher than the audio frequency, extracts characteristic parameters consisting of amplitude parameters, pitch parameters, and spectral parameters, and stores them in a data memory. In a speech synthesis device that synthesizes speech by controlling a sound source based on the read characteristic parameters, there are two cases in which a speech whose pitch changes uniformly and continuously, such as spoken words, is to be synthesized. The sound source drive cycle set based on the pitch parameter is changed when synthesizing discretely continuous sounds such as melody sounds or singing, and when playing melody sounds, the pitch signal generation The circuit generates a scale signal (approximately matching the fundamental period of the scale note) corresponding to the pitch parameter,
Since the basic cycle for driving the sound source is set based on this scale signal, it is possible to reduce the pitch deviation of the reproduced melody tones to a level that does not pose a problem for use. This has the advantage that there is no need to change the circuit configuration and bit configuration of the voice synthesis circuit such as the sound source ROM and sound source.

[Brief explanation of drawings]

Fig. 1 is an explanatory diagram of the principle of the speech synthesis method according to an embodiment of the present invention, Fig. 2 is a block circuit diagram of the same as the above, Fig. 3 is an explanatory diagram of the same as the above, and Figs. 4 to 6 are the main parts of the same as the above. The circuit diagram and FIGS. 7a and 7b are explanatory diagrams of the same operation. M is a data memory, 6 is a sound source ROM, 17 is a matching circuit, 18 is an address counter, 19 is a sound source,
31 is a scale signal generation circuit, 31b is a sound source ROM,
31c is a preset counter, 32 is a reset pulse generation circuit, 33 is an audio-melody switching circuit,
34 is a melody control code detection circuit.

Claims (1)

[Claims] 1. An audio signal is sampled with a sampling pulse having a frequency higher than the audio frequency, characteristic parameters consisting of an amplitude parameter, a pitch parameter, and a spectrum parameter are extracted and stored in a data memory. In a speech synthesis device that controls a sound source and synthesizes speech based on the read characteristic parameters, the value of an address counter that reads sound source data from the sound source ROM by counting synchronization pulses with a period equal to the sampling pulse is A matching circuit that outputs a matching signal when matching a pitch parameter, and a scale that reads out stored scale data using address data corresponding to the pitch parameter.
A scale generation circuit consists of a preset counter that takes ROM and scale data as preset input and subtracts clock pulses with a higher frequency than the synchronization pulse, and a synchronization pulse that is generated immediately after the scale signal is obtained from the zero detection signal of the preset counter. It is equipped with a reset pulse generation circuit for output and a voice-melody switching circuit for switching the reset signal of the address counter to the output of the reset pulse generation circuit or the output of the matching circuit, so that the pitch of the sound changes uniformly and continuously like spoken words. When synthesizing a continuous voice, reset the address counter using the output of the matching circuit. When synthesizing discrete sounds such as melody sounds or singing, reset the address counter using the output of the reset pulse generator circuit. A speech synthesis device characterized in that sound source data is repeatedly used. 2. A melody control code detection circuit is provided to detect the melody control code added to the amplitude parameter, and the output of the melody control code detection circuit is used to detect audio.
2. The speech synthesis device according to claim 1, wherein the melody switching circuit is controlled.