JPS59128600A - Voice synthesizer - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

[Detailed Description of the Invention] [Technical Field] The present invention relates to a speech synthesizer made of a one-chip LSI, which can be incorporated into various electrical products such as a Matusage chair, an audio alarm clock, an audio alarm, and an audio time signal device. It is used for purposes such as outputting voice messages.

[Background technology]

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 of the sound, or the fundamental period, and the The spectral parameters that represent the timbre or spectral distribution of
(hereinafter abbreviated as S-parameter). Therefore, to synthesize speech, the speech signal is sampled with a sampling pulse having a frequency sufficiently higher than the speech frequency.
It is sufficient to extract each characteristic parameter and store it in advance in a data memory, and to synthesize speech by driving a sound source based on the characteristic parameters read from the data memory.

In this type of speech synthesis device, the greater the number of samplings of the speech signal, the more faithful the speech can be synthesized.However, on the other hand, as the number of samplings increases, the number of hits in the speech synthesis data increases, resulting in a large amount of data. This requires more memory, complicates the data processing circuitry, and increases costs.

Therefore, in conventional speech synthesizers, the sampling pulse frequency (hereinafter referred to as sampling frequency) is set to the minimum frequency required to faithfully reproduce the human voice, and the sampling pulse frequency (hereinafter referred to as sampling frequency) is set to the minimum frequency required to faithfully reproduce the human voice. is 8 or 10K
Hz (sampling period 125μs or 100μs)
Set to . By the way, the audio signal is sampled using a sampling pulse to extract feature parameters consisting of A, P, and S parameters and stored in a memory. When reading and synthesizing audio using synchronous J\rus, the fundamental period of the audio reproduced based on the P parameter can only take discrete values determined by the sampling frequency. In other words, if the sampling period is 100 μS and the P parameter is Pi (an integer value), then the fundamental period to be reproduced [is t = 100Pi x 10 (sec) (where P1
The audio frequencies that can be reproduced as 1.2.3...) are discrete values.

Even if only such discrete audio frequencies can be generated, human voices and the like can be reproduced with relative fidelity. However, when playing a melody sound composed of scale frequencies, the scale and frequency of each scale (do, shi, 3) are
- There are many things that are not included in the above-mentioned discrete values, and there is a problem that if a melody sound is reproduced using such discrete audio frequencies, a melody sound with a significantly shifted pitch will be reproduced.

Therefore, as shown in Japanese Patent Application No. 56-69950, when the pitch of the sound changes depending on the scale frequency, such as in me0:p and singing, a separate scale pulse generator is used. When synthesizing speech in which the fundamental period of the sound source is determined by a circuit and the pitch of the sound changes uniformly and continuously, such as spoken words, the discrete fundamental period t determined by the above formula is determined as the fundamental period of the sound source. A melody code detection circuit for detecting a melody code from input parameters has been developed to distinguish between melody sounds and voices. It is necessary to provide this in the speech synthesizer, requiring an extra chip area, and there is a problem in that it is necessary to store an extra melody code in the data memory. Furthermore, if such a melody code detection circuit is omitted, there is a problem in that one additional input pin is required to distinguish between melody sounds and voices.

As a completely different conventional example, as disclosed in Japanese Patent Application Laid-Open No. 52-28211, a sound source ROM storing sound source data for a male voice and a sound source ROM storing sound source data for a female voice are switched. A speech synthesizer has been developed for this purpose, but in order to switch between such sound source ROMs, one additional sound source selection input pin for selecting the sound source ROM is required. By the way, in this conventional example, in addition to the sound source data of the male voice and the sound source data before the female voice, the sound source data of the melody sound is also stored in another sound source ROM.9. If the melody sound can be switched and used, it should be possible to make the synthesized melody sound a beautiful sound, but in this case, a melody code detection circuit should be provided to distinguish between the melody sound and the voice, or a melody code detection circuit should be provided to distinguish between the melody sound and the voice. It was necessary to provide one extra input pin to distinguish it from audio.

[Purpose of the invention]

The present invention was made to solve the above-mentioned problems, and does not require a melody code detection circuit.
It is an object of the present invention to provide a speech synthesis device that can selectively use any two of pre-male voice, female voice, and melody sound using one input pin.

[Disclosure of the invention]

(Structure) FIG. 1 is a so-called claim correspondence diagram showing the structure recited in the claims of the present invention.

As shown in the figure, the present invention samples and links an audio signal with a sampling pulse having a frequency higher than the audio frequency, and extracts the amplitude parameter A, the pitch parameter P, and the spectral parameter Si: In a speech synthesis device consisting of a 1-chip LSI that controls a sound source and synthesizes speech based on characteristic parameters read out from a data memory (stored in 21tal ~ data memory fl) +2+ +3). ,
The first and second sound source ROMs are counted by counting synchronizing pulses CK with the same period as the ringing pulse (4) (5)
an address counter (6) for reading sound source data from the
a matching time wKr (7) that outputs a matching signal when the value of the address counter (6) matches the pitch parameter P;
A scale pulse generation circuit (8) that generates a pulse with a scale frequency corresponding to the pitch parameter P, and a set pulse generation circuit that outputs a synchronization pulse CK immediately after the pulse output from the scale pulse generation circuit (8) is obtained. (9), a switching circuit (lO) that switches the reset/sort signal of the address counter (6) to the output of the reset pass generation circuit (9) or the output of the matching circuit (7), and the sound source selection input pin (
11), the first and second sound source ROM +
4)+5) is provided, and when the first and second sound source ROMs are mask-stored in each sound source data of the melody sound and the voice in the 6+, K is j\! Connect the sound source selection input pin (lt) to the switching input of the switching circuit (lO) so that the switching circuit (lO) is switched to the reset pulse generation circuit (9) only when sound source ROM [4] is selected. , when both the first and second sound source ROMs (4) and (6) have mask-stored audio sound source data, a trap is provided so that the switching circuit (10) is always connected to the output side of the matching circuit (7); Switching circuit (10
), the mask pattern of the switching input can be selected.

In the speech synthesis device of the present invention having such a configuration,
One of the following three mask patterns (I) to Ql is selected and used.

(+) When the first sound source Ro M I+) stores the sound source data of the melody sound as a mask, and the second sound source ROM (51) stores the sound source data before the male voice. In this case,
Connect the sound source selection input pin (11) to the switching input of the switching circuit, reset the address counter (6) by the output of the set pulse generation circuit (9) when the first sound source ROM (4) is selected, When the second sound source ROM (5) is selected, the address counter (6) is reset based on the coincidence detection output of the coincidence circuit (7). As a result, when synthesizing melody sounds, it is possible to output a sound with a pitch that matches the scale frequency, and when synthesizing a male voice, it is possible to output sounds with a discrete fundamental period that is not related to the scale frequency. .

(II) The sound source data of the melody sound is mask-stored in the IGl sound source ROM [4], and the second sound source ROM (a)
In this case, the sound source selection input pin (11) is connected to the switching circuit (10).
is connected to the switching input of the first sound source ROM (
When selecting 4), the address counter (6) is reset by the output of the reset pulse generating circuit (9), and when selecting the second sound source ROM +5), the matching circuit (7) is reset.
The address counter (6) is reset by the match detection output.

(lit) When the first sound source ROM (f+) stores the sound source data before the male voice as a mask, and the second sound source ROM (51) stores the sound source data for the female voice as a mask. In this case, the switching circuit (lO ) switching input and sound source selection pin (!
1) is cut off, and the switching input is grounded. As a result, the switching circuit (lO) is always connected to the output side of the minus number circuit (7).

Therefore, in this case, the scale pulse generation circuit (8) and the reset pulse generation circuit (9) are not used.

In the present invention, by selecting and using any one of the three mask patterns (1) to (III) above, (1) the male voice front and the melody sound are input to the input pin. (II) A voice synthesis LSI that can switch between female and melody sounds using an input pin, and (III) A voice synthesis LSI that can switch between pre-male and female voices using an input pin. One of these can be selected, and when melody sounds are synthesized, the address counter (6) is automatically reset and the G signal is switched to output a melody sound that matches the scale frequency. It is something that can be done.

(Embodiment) FIG. 2 is a block diagram of a PARCOR type speech synthesis device according to an embodiment of the present invention. As shown in Figure 3, the PARCOR type voice synthesis method samples the voice signal (Vs) with a sampler J\Rus at an appropriate period (to), and the sampled signal is between Xi and Xl-p ( The PARCOR coefficient (partial autocorrelation coefficient: hereinafter abbreviated as parameter) obtained by excluding the correlation due to the P-1> sun link values and extracting only the correlation between Xt and xt-p is used as the S parameter. It synthesizes audio, and the K parameter is the audio signal ff5 every appropriate period (to) (approximately 100 μsec) in one frame (5 to 20 m5 ec) where the audio can be considered to be in an almost steady state.
), the correlation coefficient between adjacent sample values is set to 1, and the influence of sample values sandwiched between them is set to a minimum of 2 between sample values that are separated by multiple intervals.
Find the correlation coefficient by linear prediction using the multiplicative error and subtract them! +o is added to ~. In this case, the parameter is 1, and the coefficient representing the partial autocorrelation with points near XtK, such as Kx - Ks, contains a wealth of information regarding the spectral distribution,
Partial autocorrelation with points far from Xt, such as Kso, does not contain much information about the spectral distribution, so a large number of quantization pits are assigned to the low-order parameters, and a large number of quantization pits are assigned to the high-order parameters. It is more effective to allocate a small number of quantization pits to reduce the number of pits and reduce redundancy. Therefore, the PARCOR method has a better band compression rate than the autocorrelation coefficient method, which uses an autocorrelation coefficient as an S parameter and allocates the same number of pits to each coefficient.

Normally, the parameters for each A, P, and P are compressed and stored or transmitted, and each coefficient KISK is 5 pits for the A parameter, 6 pits for the P parameter, and 6 coefficients for the K parameter.
7.6.5.4.4.4.3 for z...K1°
3.3.3 Assign as Pitt et al.

The pARcOR type speech synthesis device shown in Fig. 2 includes a control IC (A) equipped with a data memory for storing voices and melodies as compressed feature parameters, and a speech synthesis IC (dotted line areas A and B). (excluding the part) and
Data is transferred between both ICs in a pit serial manner. By the way, all voice characteristic parameters are stored in the playback ROMH as 10 pit data, 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. has been done. For example, in the case of All parameters, 32 pieces of data expressed by 10 pits are stored. Therefore, the number of relative address hits required when accessing arbitrary data of the A parameter is 5 hits. This relative address is called a compressed parameter because it represents the characteristic parameter compressed to the minimum necessary size. On the other hand, RoMH for playback
The actual feature parameters stored within are called playback parameters. As is clear from the above, the number of hits of the playback parameters is 10 pits in common for each characteristic parameter of A-P, , Kt to Kxo.
The number of pits of the compression parameter is different for each parameter of A, P, K1 to KIO, for example, 5.6.3.3S3.3.4.4.4.5.6.7 pits (53 pits in total). ). Such a compression parameter is 5 to 29 m5, which allows the audio signal to be considered to be in an approximately steady state.
Since one set (-53 hits) is extracted for every ec (l frame), the audio signal can be reproduced by processing the data at a rate of at most 2650 hits/second, and even silent sections and reheat sections can be reproduced. Taking this into consideration, it is actually possible to reproduce audio signals at about 1600 pits/second. By the way, in the embodiment, the fundamental period generation is explained in the case of synthesizing speech in which the pitch changes evenly and continuously, such as spoken words, and in the case of synthesizing speech that continues discretely, such as melody sounds and singing. The method has been changed, and when reproducing melody sounds, the melody sound is synthesized by driving the sound source with a basic cycle equal to the basic cycle of each scale note.

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., the relative address of the playback ROM θ3) are stored pit-serially from the data input terminal θ through the switching circuit θ→ in the ring register 0 for each frame. However, since the reproduction data of each parameter is stored continuously in the reproduction ROM (+3), it is not possible to extract specific data. Therefore, the index ROM (17
) The starting address of each parameter in the playback ROM (13i) stored in the playback ROM (13i) is sequentially taken out under the control of the address counter (18) and added to the above relative address by the adder circuit 09. ROM (
+3) absolute address (9 hits) is calculated, and the playback ROM (13) is accessed using this absolute address. The index ROM Q7) stores the number of pits allocated to the compression parameter as a 3-pit binary number, and data regarding the number of pits allocated to the compression parameter is sent to the reproduction control circuit Kaoru0), which controls the regenerator.
circuit! 20) sends shift clocks as many as the allocated number of pits to the ring register (+6). Therefore, the ring register (1→→...) corresponds to the above pit distribution number, for example, 5 pits for the A parameter, 6 hits for the P parameter, 3 hits for the 10% parameter, etc. . . . In the case of K+ parameters, compressed parameters (relative addresses) such as 7 hits are sent serially to the adder circuit θ9). In addition, the index R'OM (the first address in the reproduction ROM (+3) of each characteristic parameter stored in lη is sent out one pit at a time to the addition circuit (19) via the 1-parallel-serial converter circuit 1). Because it is done,
The absolute address is calculated by sequentially adding each hit one by one. The serial absolute address thus calculated is converted into parallel data via a serial-parallel conversion circuit and converted into an address for accessing the reproduction ROM (13).

The feature parameters output from this playback ROMOS are updated every frame, but if the feature parameters change discontinuously at the connection point between each frame when updating data, the audio signal will be distorted. Therefore, an interpolation calculation circuit (23) is provided to perform approximate linear interpolation at 8 points within one frame so that feature parameters can change smoothly when updating data. I try to do this. For this reason, the tinning control circuit (in the case of 2 Shu, 8 interpolation D clocks (2
, 5m5ec), and 25 parameter read P resets (100μsec) during one D clock.
Furthermore, 22 bit reading T clocks (4.5 μsec period) are created during one P clock. Note that the P clock is a synchronization pulse corresponding to a sampling pulse. Data is read into the ring register (+6) from the data input terminal 0→ at the first Dl of the eight D clocks. Each compression parameter A, P, KL. ...
, K1 are read sequentially at odd-numbered P clocks. For example, the A parameter is read from T6 to Tl in the P1 interval.
The 0th even-numbered PO which is read by the 5 T-tricks of o or the T-tricks other than the above are processed by the interpolation calculation circuit (
This is used as the timing for the sound source ROM +41 +5), digital filter, etc.

This interpolation calculation circuit (4 stops its operation when melody sounds are synthesized; M is a parallel-to-serial conversion circuit).

Each feature parameter updated to a new value every 2.5 m5ec by the interpolation calculation circuit is temporarily stored in the P latch (27) and the AK latch, respectively. However, all parameters that are not required for the time being for interpolation calculations are transferred to the AK parameter stack I29) and stored as data for speech synthesis in the digital filter.

By the way, the P parameters stored in the P latch are data used to drive a sound source to generate an impulse signal having a fundamental period corresponding to the P parameters. Sound source ROM that counts P clocks equal to pulses +
The reset signal of the address counter (6) at 4+ +61 becomes the output of the matching circuit (7) that detects the match between the output of the address counter (6) and the P parameter stored in the P latch, and the address counter (6) The voiced sound source control data based on the P parameter is output from the sound source ROM +4+ +5), which is reset at a cycle that is an integral multiple of the clock cycle (P parameter), and is sent via the sound source switching circuit (12). Input to the switching circuit (to). When the sound has no fundamental period, the switching circuit is driven by the sound source control circuit to switch to the unvoiced sound source (c), and the power of the unvoiced sound source is the white noise (white noise). The next KA parameter and the second parameter are supplied to a digital filter, which reproduces the sound by adding information regarding amplitude and spectral distribution to the signal supplied from the sound source. Color 3) in the figure is a low frequency amplifier that amplifies the reproduced audio signal, - is a speaker,
The side is a crystal oscillation circuit.

Scale pulse generation circuit (8) shown in Figures 5 and 6 below
, the configuration of the reset pulse generation circuit (9) and the voice synthesis operation for synthesizing melody sounds will be explained. The scale pulse generation circuit (8) is connected to a shift register (36) in which the data corresponding to the P parameter, that is, the compressed P parameter outputted from the control ICm, is input by a request signal (VRE), and the compressed P parameter is input to the address data. The L7 scale ROM (3η and the scale ROMe
The desire level data read from η is used as a preset input and P
It consists of a preset counter (9) that counts the 0 clock pulse, for example, the T clock, which has a higher frequency than the 0 clock pulse, and an inverter 9 that inverts the t!o detection signal of the preset counter cup. A gap detection signal having a cycle that is an integral multiple (scale data) is output as a scale pulse (PM). In this case, the frequency of the scale pulse (PM) output from the scale pulse generation circuit (8) takes discrete values, but the discrete interval becomes smaller in accordance with the frequency of the zero pulse. Therefore, by storing appropriate scale data in η of the scale ROM, the scale pulse generation circuit (8) can generate a scale pulse (PM) that matches the frequency of each scale. For example, if the 0 pulse is set to 0 (period: 4.5 p sec)
) and set the compressed P parameter corresponding to the P parameter "12" so that the scale data "284" is read from the scale ROM (3η), the preset counter (to)
A zero detection signal is obtained with a period of 4.5 × 284 μsec, and this scale pulse (PM) is
This means that a discrete fundamental period corresponding to the P parameter can be interpolated with a fundamental period corresponding to 12.8J. The reset pulse generation circuit (9) is an inverter AOI (41),
The capacitor is formed by four lines: a Nant gate (to), a D flip Florazu (to), and an and gate (to), and the scale signal output from the preset counter (c) as shown in the time chart in Figure 7. The P pulse immediately after obtaining (PM) is output as the reset pulse (VR) of the address counter (6). In the figure, (a) shows the pulse when the P parameter is "12". The output of the matching circuit (7), (b) indicates a scale signal (PM), and 0 indicates a reset pulse (VR).

The first sound source ROM f4) in which the melody sound is currently stored in the sound source selection input pin (11) from the control IC (A)
When a signal is given to select , the switching circuit (lO) is switched to the set pulse generating circuit (9) side) and the counter (6) is output from the reset pulse generating circuit (9). The address counter (6) is reset by the reset pulse (VR)
The number of cases in which the number is counted and then reset occurs at a ratio of 4:1. Therefore, the sound source ROM is
+4) is accessed, and the scale note "G" is accurately reproduced. In the same way, each scale note is played accurately,
The melody is played at the correct pitch.

The time chart shown in Figure 8 explains the relationship between the scale pulse (PM) and the reset pulse (VR) more easily.
The reset pulse (VR
). As is clear from the figure, 3S6.8-11.1 of the P pulse is used as the reset pulse (VR).
4.16th pulse is output. The address counter (
00 of the sound source ROM (4) 5% addressed by 6), isometrically 3.75 K H from the sound source ROM (4)
If the sound source data is read out at a period that can be considered as z (3 μsec), the sound source will be driven at the correct scale frequency, and sounds such as melody sounds and singing will be reproduced at accurate pitches.

However, in Figure 2-099, the sound source selection input pin (11) is connected to the switching input of the switching circuit (lO),
When selecting the first sound source ROM (+4), the switching circuit (10
) is switched to the reset pulse generation circuit (9) side, and when the second sound source ROM +5+ is selected, the switching circuit flo) is switched to the matching circuit θη side, and the first sound source ROM +4) is switched to the matching circuit θη side. The problem is that the sound source data of the melody sound is stored as a mask, and the second sound source R It is also possible to ground the switching input of the switching circuit (10) by changing the mask pattern in the area surrounded by
) is connected only to the coincidence circuit (7) side, and the scale pulse generation circuit (8) and reset pulse generation circuit (9) are not used. In this case, the first and second sound source ROMs +4+ +5) mask the sound source data of male and female voices, respectively. Note that such sound source data is data indicating the waveform of a residual signal obtained after inputting a male voice sound, a female voice sound, a melody sound, etc., respectively to the PARCORJI voice analysis filter and extracting the A parameter and the 2 parameter. , sound source ROM (4) (5) that can store such residual waveforms.
) is reset at regular intervals (6
), a residual signal having a fundamental period can be reproduced by repeatedly accessing it.

〔Effect of the invention〕

As described above, when each sound source data of melody sound and voice is mask-stored in the first and second sound source ROMs, the switching circuit is reset only when the first sound source ROM is selected. When the sound source selection input pin is connected to the switching input of the switching circuit so as to switch to the side of Since the mask pattern of the switching input of the switching circuit can be selected as shown in FIG. When the sound source ROM is selected, the reset signal of the address counter can be switched by the switching circuit. It can be switched in conjunction with the ROM selection, and therefore has the effect of synthesizing playback sounds with accurate pitches when synthesizing melody sounds, and when creating a voice synthesis LSI that can select between male and female voices, the sound source By separating the selection input pin from the switching input of the switching circuit, it is possible to avoid using a scale pulse generation circuit, etc., and furthermore, such selection can be made without using a melody code detection circuit as in the past. There is an effect.

[Brief explanation of the drawing]

Fig. 1 is a so-called claim-corresponding block diagram concisely showing the configuration shown in the claims of the present invention, Fig. 2 is a block diagram of an embodiment of the present invention, and Fig. 3 is an explanation of the principle of the same. FIG. 4 is a time chart of the 0-pass pulse used in the above, and FIG. 5, FIG. 7, and FIG. 8 are operation explanatory diagrams of the same. tx) +21 +3) is data memory, +4) +5)
is the sound source ROM, (6) is the address counter, and (7) is -
Several circuits, (8) is the scale I\rus generation circuit, (9) is the reset pulse generation circuit, (lO) is the switching circuit, (II) is the sound source selection input pin, 02 is the sound source switching circuit, and CK is the synchronization pulse. be. Agent Patent Attorney Ishi 1) Choshichi

Claims (1)

[Claims] +11 Sampling the audio signal with a sampler pulse of a higher frequency than the audio frequency, extracting characteristic parameters such as amplitude parameters, pitch parameters, and spectral parameters and storing them in data storage,
A 1-chip LS that controls the sound source and synthesizes speech based on the characteristic parameters read from the data memory.
In a speech synthesis device consisting of I, a matching device outputs a matching signal when the value of an address counter that reads out sound source data from the first and second sound source ROMs by counting synchronizing pulses with a period equal to the sampling pulse and matches the pitch parameter. circuit, a scale pulse generation circuit that generates a pulse with a scale frequency corresponding to the pitch parameter, a set pulse generation circuit that outputs a synchronization pulse immediately after the pulse output from the scale pulse generation circuit is obtained, and an address counter. a switching circuit that switches the reset signal of the RO to either the reset pulse generation circuit output or the matching circuit output; and the switching circuit connected to the sound source selection input pin,
A sound source switching circuit for switching the first and second sound sources is provided.
When each sound source data for melody and voice is mask-stored in the sound source ROM, the sound source selection input pin is connected to a switching circuit so that the switching circuit is switched to the reset pulse generation circuit only when the first sound source ROM is selected. When connecting the switching input of the switching circuit to the switching input of the switching circuit and masking the audio sound source data in both the first and second sound source ROMs, connect the switching circuit to the matching circuit output side at all times. A speech synthesis device characterized in that patterns can be selected.