JPS58158696A - 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]

The present invention compresses feature parameters extracted by suitably processing an audio signal and stores them in a data storage unit, and synthesizes audio based on the compressed parameters read from the data storage unit. The present invention relates to a speech synthesis device, and its purpose is to provide a speech synthesis device fk that can selectively reproduce a plurality of types of voices with different volumes or pitches corresponding to each compression parameter. In general, the amplitude parameter (hereinafter abbreviated as A) which expresses the magnitude of the sound by linking the audio signal with a siren pulse with a frequency higher than the audio frequency, and the basic Pitch parameter representing the period (
Extracting characteristic parameters consisting of a spectral parameter (hereinafter abbreviated as an S parameter) representing the timbre or spectral distribution of a sound (hereinafter abbreviated as a P parameter). Each feature parameter is compressed into a T number of bits according to the degree of contribution to sound quality, stored in a data storage unit at 1° with the compression parameter, and read out sequentially from the data storage unit. Access the playback ROM in which the parameters are stored, and use the characteristic parameters continued from the IL OM for 11th grade students to advance the source and play back the entire audio. At the same time, it is necessary to store in the data storage unit 1 compressed harashita corresponding to each line i or pitch f.
child. 1. Will the volume or pitch be adjusted according to the surrounding noise condition or the user's preference? In order to be able to read the voice of each volume or pitch, it is necessary to store each compression parameter in the data storage unit, and the memory valley M of the data storage unit is more than necessary. The drawback is that the P/IRcOR type sound angle synthesizer has to be made larger.The present invention has been designed in view of the above drawbacks.Hereinafter, an example of the P/IRcOR type sound angle synthesizer will be described. 1 and 1 will be explained by ;tJ.The COR type sound synthesis system for PA is shown in Figure 1.''--The VC audio signal (Vs) is sampled (-1) at an appropriate period (to) by a gas printer pulse. ``fyu'J nk large f1.r, :+1'
(P) between xt -p)
-1) Excluding the correlation due to the number 11,
X

Extract all the correlations between [and Xt-p and 1 piece PARC
Speech is synthesized using the OR coefficient (partial autocorrelation coefficient: hereinafter abbreviated as parameter) as the S parameter, and the K parameter is used for one frame (5 to 20 m sec,) in which the speech can be considered to be in an almost steady state. , the sound is Ji+(■
S)'s sunburi, Jri complete line, adjacent sunzuri,
The correlation coefficient of the 7-value opening is set to 1, and the isolation between multiple values t'IT
Between Kosanzuri 2i evening price, Kyouma 711 Kosa'
The influence of the J-Rari y-g value is determined by linear prediction using the least squares error, and the correlation coefficient obtained by subtracting the values is 1 and the correlation coefficient is 1, and the parameter is Kl. 2, 1 to 3, the coefficients representing the partial autocorrelation with points near xty (H) include the spectral distribution.
Partial autocorrelation coefficient f with points far from K[ such as KIO
Since it does not contain much information about the spectral distribution, a large number of magnetization hits are applied to the low-order parameters, and a small number of magnetization hits are applied to the high-order parameters.
It is more effective to reduce the cost of redundancy by allocating resources. However, this method has a better band compression rate than the autocorrelation coefficient method in which the PARCOR method uses the autocorrelation coefficient as the is parameter and allocates the same number of pits to each coefficient. Normal valley A. The parameters are compressed and stored in P, and then transmitted.
5 hits for the A parameter, 6 pits for the P parameter, and each coefficient Kl of the K parameter. K2... Paired with Klo (2, 7, 6, 5, 4,
4, 4, 8. Assign K i & l J like S, S, Sji, Soto, etc. The following is a description of the entire illustrated embodiments of the present invention--embodiments.
r Ill will explain. FIG. 3 is a 099 diagram of the speech synthesis device d according to the present invention. As shown in the figure, this speech synthesis device includes a control IC (A
) and a voice synthesis IC (dotted line areas A and B with 1 extra), and is composed of 2,000 blocks, and it is a 1-2 block so that data is exchanged between the two in a pit serial manner. be. All audio feature parameters are stored as 10 bits of data in the playback ROM, and the number of data assigned to each feature parameter is optimally distributed according to the degree to which that feature parameter contributes to sound quality. has been done. Funeral diagram 4 is stored in the playback ROM ill.
, A, P, KIO to Kl. For example, in the case of the A parameter, 32 pieces of data expressed by 10 pits are recorded. Therefore, there are five relative address pits required when accessing any number of A parameters. This relative address compresses and expresses the feature parameter to the necessary minimum and is called a compressed parameter because it is one piece. This fl,
[On the other hand, the actual characteristic parameters stored in the playback ROM H are called playback parameters. Above [7
From here, set the playback parameter pit @ to A, , P, KIO ~ to 1&Ml”f
ly 7~ All parameters are common [10 pits, but the number of pits in the compression parameter is A. P, K1o~ are different for each parameter of 1, fl, 5.6.8, 8, 8.8, 4, respectively.
4°4.5,6+7e points (53 pits in total). In addition, there is a reserve area for 3 bits, that is, data 8.
A portion for each is reserved in the playback ROM. The compression parameters are 1 for every 20 rn sea (I frames), which can be considered as an almost constant K state.
Since pairs (=53 pits) are extracted, d26
It is possible to record all voice messages at 50 ps/sec, and it is also possible to record silent sections and reheat sections.
(is capable of recording an audio signal at approximately 1600 pits/second.) Based on this, the compression parameters (i.e., the relative address of the playback ROM II+) specified in data record 1 (8) ) V=t + switching circuit for each frame (1
0) to the link register (3) via a relative address t like this.
Since it is not possible to retrieve the stored data from the playback ROM t++ due to the cracks, the control of the first address full address calculator (11) is stored in the memory card OMf21 as shown in Figure 5. Take it out one by one from below,
By adding the relative address and the addition circuit (4), the absolute address (9 hits) of the reproduction ROM to is calculated, and the absolute address is used to calculate the absolute address of the reproduction ROM
I'm trying to access it from file to file. R for reproduction below
The previous readout of the playback harameter stored in the OM 11) will be described in detail. The index R01vj [2] stores the number of pits allocated for compression parameters as a 3-pit binary number, and has 1 pit for commonization of T to reduce the storage capacity of the playback ROM full. i(OM (A reserve reserve corresponding to the reserve area of 11 yen is provided. Hit distribution of compression parameters @
The data regarding the regeneration control circuit (12) is not sent to the field, but the reproduction control circuit (12) sends all the shift clocks corresponding to the number of pit allocation to the shift register (3). 3 (from, “2sI
For example, in the case of A parameter, it is 5 pits, and in the case of P parameter, it is 6 pits, K.
t. In the case of a parameter, it is compressed to 3, υ, 1 (in the case of an l parameter, it is compressed to 7), and the parameter (relative attribution) is sent serially to the JJII world circuit. Yes, :, Il) The register (3) is constructed of a dynamic shift register so as to minimize the burden on children. The leading address in the ROM (1) for reproducing each characteristic parameter stored in the mafko index ROM (21) is sequentially added by 1 and 1 through the parallel/serial switching circuit (13). 4), the absolute address is calculated by sequentially adding one bit at a time.The calculated one bit is converted into an absolute address of the serial data and converted into parallel data via the serial/parallel converter (14). The playback ROM (1) can be fully accessed. (9) in the figure is a parameter code detection circuit. This characteristic parameter is read from the playback ROM (1) and is sent to the volume correction circuit (Boa) as appropriate. ) and pitch correction circuit (80b
) K is inputted to the interpolation calculation circuit (5) via I, and the pitch correction circuit (30a) adds unreasonable pitch correction data to the A parameter of the feature parameters. is subtracted and outputs the corrected A harammeter, -10,000,
The back press correction circuit (80b) adds or subtracts appropriate back press correction data to the P parameter and outputs a corrected P parameter. The specific configuration and operation of the image correction circuit (8(D) (80b) will be described later).By the way, the compensation 1?1] calculation circuit in which the correction A parameter, correction P parameter, and the parameter are input. is updated every frame to prevent audio signal distortion (decrease in intelligibility) due to discontinuous changes at connection points between frames of feature parameters.
Approximate linear interpolation is performed at 8 points within one frame so that the 4-character parameters can change smoothly. , this interpolation calculation (b) path (5) is
, and a timing control circuit (28), and a timing control circuit (28) as shown in FIG. 8 interpolation marks during vCl frame (20m kick)
J Ri (2,5m sec)'! i knee'r live,
1) P lock for reading 25 parameters (100p 1iec), and 11
1I! 'r for reading the pit of 221 hard during J's P lock
9 [1”J ri (4,5 p st, c) was created l
Ru. In the first DIVC of the 84-wire D/O check, data is transferred from the data input terminal (8) to the link register (3). Each compression parameter A, P, Lto..., K1
are read sequentially at odd-numbered P clocks, and for example, the A parameter is read at five intervals from 1'6 to rto of the P1 section tooth. Even numbered Pri0
The '1' and '0' blocks other than those mentioned above are used as q-tie and l:y-tie, such as the complementary lTh1 total I circuit (5), the sound source ROM +61, and the digital filter (7). The above bucket [ml calculation circuit tel K is 2.5m]
Each characteristic parameter, which is updated to a new value every 5 ec, is temporarily stored in the P latch (I6) and the AK latch (23), respectively. All parameters that are not needed for the time being are stored in the AK parameter stack (
241 and stored as voice synthesis data for the digital filter (7).
Data regarding the basic 1M period of J.E.L.A.T. voice, that is, P
The parameters are matching circuit 07) K and P 0 (100 μ
5ec) The address counter (18j) is reset to full power and is compared with the output of 1 barrel, and when the output of the address counter (18) matches the l'' parameter, the match circuit (17) resets and resets the address counter (18j). Shinq・(VR
) is output. Address counter t1g
j) The I' parameter is reset every cycle, and sound source system data is sequentially generated from the sound source ROM +6+ at this cycle. With this sound source control data, the voiced sound source (
19) All voiced sounds having a fundamental period are generated with full assistance. For example, if the P parameter is 1-25, a voiced sound with a fundamental period of 25x100μ (4001 (Z)) will be generated.The above sound source control 1 data is the residual sound obtained by frequency analysis of the original sound. This data is used to faithfully reproduce the tone by reproducing the difference waveform.On the other hand, when the voice does not have a fundamental period, the VC is controlled by the sound source control circuit (20). 41). Unvoiced sound source [21] This source generates white noise with i fundamental period and less than 1. Next, the A-harameter and parameters are supplied to a digital filter (7) equipped with a VCA, supplied from a sound source circuit (outputted from a voiced sound source (19) or an unvoiced sound source 121), and then converted into a signal with an amplitude of By adding information regarding the size and Sj vector distribution, the entire audio is reproduced. In addition, in Fig. 3, (25) is an amplifier, (z6) is a switch, and (271 is a Mizuhina oscillator circuit), but these are not directly related to the gist of the failure ll'!IJ. Therefore, detailed explanation thereof will be omitted.The specific circuit configuration and operation of the volume correction circuit (80a) will be explained below.
) in the figure, (81a) (81
b) is a seventh-key changeover switch (Il) to (■7 ) is an inverter circuit, (NAI) ~ (NA5)ll-1' nacido circuit,
(At) to (A4) are AND circuits, (Not) to (NO
s) is a NOR circuit, (El) (E2) is an exclusive OR circuit, (F) is a flip D for carry + P Lee generation.
・Tsubu, (AD) is 1 point, (3 forces are the correction signal generator, (33) is short arithmetic system (I14), (
34) is the addition/subtraction operation Kls. In addition, characteristic parameters consisting of serial data are read from the playback ROM (31) and are input sequentially to the input terminal (IN) in synchronization with the T lock, and A parameters are read to the tree (CO). The control clock (P2) that becomes "1" is input when the clock is in the state of 1.The clock input terminals (CLt) to (CL4) receive lock signals (TC+) to (TC7) as shown in FIG. ) is input, and the correction signal generator (32) generates the subtraction signal (VD) and correction data ra, J as shown in the figure.
A correction signal (Vs) (Va) corresponding to r6J is generated. However, - the above signal (VD) (Vs) (Vs) is generated only when the control signal is 0 (P2) or 1.It is generated in this way and the signal (VD) and correction are The signal (Vs) (Va) is controlled by the arithmetic control section (3) controlled by the t-mode changeover switch (81a) (Blb).
31, it is input to the addition/subtraction calculation section (3) as an arithmetic control input or addition/subtraction input data,
In D), the carry operator series output from the flip-flop is added or subtracted from the A parameter of the characteristic parameters read from the playback ROM ill and the volume correction data, and the result is output. Correction A to terminal (OU'r)
Parameters are output. The relationship between the 7-mode switching signal (VMI) (VM2) manually input by the t-mode switching switch (31a) (C31b) and the correction data value ΔA of the A parameter is as shown in the table below. The correction A parameter obtained by adding or subtracting the correction data value △A to the t'lTkoA parameter read from the reproduction ROM file based on the compression parameter of the data storage section (8) is calculated by the oil level calculation circuit (5). ), and the audio of each mode will be played. In addition, in the embodiment, by fully switching the two 7-mode changeover switches (81a) (3tb), a 2-hi-sight t-mode switching signal is generated, which corresponds to one compression parameter. Although it is possible to reproduce 4E-do sounds with different volumes, if a multi-pit t-do switching signal is used, more t-dos can be set. Well, A parameter correction data l+i! (
△A) is optional (it goes without saying that it can be set). Note that the specific structure and operation of the pitch correction circuit (Bob) are exactly the same as those of the volume correction circuit (80a), so a description thereof will be omitted. However, when the P parameter is read out to the control N4 (CO), a control voltage (P4) that becomes "1" is applied.
Similar to the volume correction circuit (80a), from 80b), suitable pitch oil stop data is added to supplementary IEP parameters for T-co bass 1-C, and correction P parameters or corrections for treble t-C from which pitch correction data is subtracted. The standard mode P parameters that are not used are outputted as a waste. Figure @8 shows another embodiment, including the volume correction circuit (raoa) and the pitch correction circuit (80b) of the embodiment described in B.
The signal generation for correction is formed using a full programmer zero thick array (PLAt) (PLA 2), and the calculation 11JIJ m1 capital (33) and addition/subtraction calculation section (circle is shared to achieve sound delivery, pitch stoppage, etc.). The circuit (30) is composed of the circuit (30). In the figure (DC) is the 4-bit switching signal (VM
) to a 16-pit signal, (ps) is a p-s converter that converts parallel data to serial data, (Onx) (ω2) is a control terminal, By appropriately designing the logical operation by (PLA2), it is possible to perform the same operation as in the above embodiment, and the correction data value (△
A) Setting of (ΔP) becomes easier and the circuit configuration becomes simpler. Note that a correction code for controlling volume or pitch correction is stored in the total data storage section, and this correction code is detected by the total code detection section and output as a 7-key switching signal (V
M) 'i may be generated. In addition, the pitch correction data value (△P) is +1 as 1T-do to correct the pitch.
Alternatively, set it to -1, correct the pitch for the 1-ko scale, and set the 1-do.
-), "Mi", . . . ), by providing a frame in the scale interval correction mode at an appropriate period, it is possible to completely synthesize scale tones without pitch deviation. For example, when the P parameter is "13", by setting the pitch correction for the scale by 1 do (△P-1) every 5 frames, the frame where the d voice is synthesized with the P parameter "]3" and The sound is synthesized with the P parameter "12", such as the f1 gnome, which is repeated at a ratio of 4:1, the P parameter r12, and the scale golden times (782 Hz) corresponding to 8J.
) can be synthesized. The above-mentioned scale sound synthesis method has been described in detail in the previously published VC (Japanese Patent Application No. 56-69950 and 1987-69951).
Detailed explanation will be omitted. The present invention is configured as described above, and the gear angle signal is appropriately sampled and extracted, one feature parameter is compressed and stored in the data storage section, and one is read out from the data storage section. In order to synthesize speech based on this compression parameter, the Lf speech synthesis device is provided with a volume correction circuit that adds or subtracts volume correction data as appropriate to the amplitude parameter among the characteristic parameters, and also adds or subtracts volume correction data as appropriate to the pitch parameter. A pitch correction circuit that adds or subtracts pitch correction data is provided 1. This allows the pitch correction circuit to add or subtract pitch correction data. Although it is possible to selectively synthesize multiple types of sounds with different volumes or pitches, compression parameters corresponding to each volume or pitch are stored in the data storage unit. This has the advantage that it is not necessary to store the information in advance, and the storage capacity of the data storage section can be reduced.

[Brief explanation of drawings]

FIG. 1 is an explanatory diagram of the principle of the speech synthesis method according to the present invention,
@Figure 2 is the same operation curse 1 diagram, the 3rd quotient is the improvement lock circuit diagram, and Figures 4 and 5 are the same as above for reproduction.
t (J M, Diagram showing the configuration of Ishiketsu ROM, Figure 6 is a specific circuit diagram of the main part of the improvement, Figure 7 is a diagram explaining the operation of the same as above, Figure 8 is a specific circuit of the main part of another example The figure is: (1) ILOM for reproduction, (8) redata storage section, t1
9J ull U sound source, (30a) ll-1: volume correction circuit, (3Qb) (1-1 pitch correction circuit).

Claims (1)

[Claims] (1) Link the audio signal using a link pulse with a frequency higher than the audio frequency to extract feature parameters consisting of amplitude parameters, pitch parameters, and spectrum parameters, and each feature parameter contributes to sound quality. According to the degree of compression, the compression parameters are recorded in the data storage section as compression parameters, and each characteristic parameter is stored in advance using the compression parameters read out sequentially from the data storage section. A sound source is driven to reproduce the sound using the characteristic parameters read from the ROM, and in the single voice synthesizer, appropriate volume correction data is added to the amplitude parameter among the characteristic parameters read from the reproduction ROM. Alternatively, a volume correction circuit for subtracting is provided, and a pitch correction circuit is provided for adding or subtracting pitch correction data as appropriate from the pitch parameter, and based on the corrected amplitude parameter and corrected pitch parameter output from the volume correction circuit and the pitch correction circuit. A speech synthesis device f characterized in that it is configured to reproduce speech.