JPS6139099A - Quantization method and apparatus for csm parameter - 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 CSM parameters, that is, CSM (Compoait6S) expressed by at most 4 to 6 frequencies.
i-nusoidal modeling: composite sine wave model) related to parameter quantization.

(Prior Art) Conventionally, LPC-type speech synthesizers have been widely used as speech synthesizers, but LPC-type speech synthesizers generally have a complicated structure. Also, the characteristics of the LPC filter used for speech synthesis are
There is a drawback that its stability is impaired due to errors during parameter transmission.

On the other hand, the CSM type speech synthesizer that performs speech synthesis using CSM does not have a filter and has a very simple structure, as will be explained in detail later. No stability problems arise. However, regarding the quantization of CSM parameters, conventionally, each amplitude of the parameter has been quantized separately, and the relationship between the parameters has not been considered. Therefore, quantization that fully utilizes the characteristics of the CSM parameters is not performed, resulting in a disadvantage that the efficiency of quantization is low.

(Objective of the Invention) An object of the present invention is to solve the above-mentioned problems when quantizing CSM parameters and to provide an efficient quantization method.

(Structure of the Invention) The quantization method of the present invention is based on the CSM parameter quantization in which the spectral envelope of speech is expressed as a set of sine waves with a predetermined number of free frequencies and amplitudes. A collection of (doors,
, m, ,...-1) is α=exclusive ax(m, m,,...
. . . -1) It is configured to have means for normalizing using a normalization coefficient α expressed as:

(Hara Ri) First, the principle of the CSM type speech synthesizer will be explained.

CAM expresses an audio signal as a sum of a specific number of sine waves having amplitude and frequency as freely selectable parameters. A predetermined number of 4 to 6 sine waves is used at most.

Therefore, when performing CAM speech synthesis, it is first necessary to express the speech signal as a sum of a predetermined number of sine waves by CSM speech analysis.

CSM voice analysis will be explained in detail later, and only the main points will be explained here.

In CSM analysis, as in LPC analysis, an autocorrelation coefficient is used as an intermediate parameter for the purpose of ignoring phase information, averaging the influence of sound sources, and avoiding instability due to noise components.

That is, in CSM analysis, N low-order taps of the sample autocorrelation coefficient directly calculated from the speech waveform to be expressed for each analysis frame are combined with N low-order taps of the autocorrelation coefficient of the composite wave. The purpose is to determine the frequency and intensity (power amplitude) of each sine wave to be synthesized so that they match the N sine waves.

Now, let the number of sine waves to be synthesized be , and let the angular frequency of each sine wave be ω6 (i-x, z, ..., s, and the intensity of each sine wave be m, then the CSM composite wave yt is yt=zF
Within i (ωit+φ,) 1 fog! However, the autocorrelation coefficient 46 of this tap L is ωi,
It is easily expressed using nLi, rt=, Σm@cmtc
d6 [=1.

On the other hand, if the sample of the audio waveform to be expressed is Xt, the sample autocorrelation coefficient Ut of tap t in a certain seven frames is given as follows. However, M is the number of samples in one analysis frame.

Now, in the CSM analysis, the above-mentioned rt is equal to the given tjt and the valley t
, ω, is determined.

That is, the values of door i and ωi are determined so that 't', 't', where t=0.1, 2.., N holds true.

This specific method will be explained in detail later.
Here, it is assumed that the above-mentioned sine waves 11Li and ωi are obtained one after another for each analysis frame in response to a given audio signal.

An example of the audio feature vector pattern based on the CSM parameter 1ωiK obtained in this way is shown in FIG.

In addition, the 9th order (N = 9) CSM (number of sine waves rb = 5
)-) FIG. 2 shows an example of the correspondence between the in-spectrum and the 9th-order LPC spectrum envelope (frequency transmission characteristic of the LPG synthesis filter) obtained from the same audio sample.

Note that there is a relationship between the above-mentioned order N and the number of sine waves as N==2×1 as described later.

From these figures, it can be seen that the CSM contains information extracted from the features of the original speech to be expressed.

However, using the values of mi and mi obtained as a result of the CSM analysis, we can create a set of If we simply create a sine wave and synthesize it simply and additively, the human ear simply hears the sound as a synthesized sine wave, and the goal of reproducing the original sound cannot be achieved.

This means that even if sine waves are simply added, the spectrum of the generated signal is only a discretized line spectrum, whereas the spectrum of the audio signal has a continuous spectral envelope, and First, voiced sounds are expressed by a pitch structure, and unvoiced sounds have a fine spectral structure expressed by a stochastic process.By simple addition, 90SM and speech signals have completely different spectral structures. It is thought that

Therefore, in order to synthesize speech using CSM, it is necessary to use some method to spread the line spectrum into a continuous spectrum. In other words, CSM speech synthesis can be thought of as generating a speech spectrum pattern from a speech feature vector pattern expressed by a line spectrum as shown in FIGS. 1 and 2.

In the present invention, the following techniques are used to perform the above-mentioned spectrum spreading in CSM speech synthesis.

That is, since a voiced sound has a clear pitch structure, the phase of each of the sine waves specified as described above is reset every pitch period. As a result, it becomes possible to easily generate a spectral envelope and a fine spectral structure of bis and tris.

Furthermore, by applying special time window processing to the above-mentioned phase reset waveform as detailed in the explanation of the embodiment, discontinuity of the synthesized waveform at the time of phase reset can be removed and continuity of the audio waveform can be ensured. are doing.

As a result of the above implementation, the CSM line spectrum shown in FIG. 2 is diffused as shown in FIG. Experimental results have shown that sound quality that can withstand practical use can be obtained.

For reference, FIG. 3(b) shows a CSM spectrum obtained by simple addition without performing the above-mentioned processing.

As mentioned above, a waveform with such a spectrum simply sounds like a synthesized sine wave sound, and the purpose of reproducing audio is not achieved.

The above is for voiced sounds, but in the case of unvoiced sounds, it is performed as follows. In other words, in the case of voiced sounds mentioned above, the phase reset and special time window processing performed for each pitch synchronization are performed, and in the case of unvoiced sounds, instead of pitch synchronization, the synchronization that occurs randomly as a stochastic process is distributed. Using a pulse whose width and lower limit value are set, the above-described processing is performed every time this pulse is generated.

By using the above-mentioned method, it is possible to perform CSM synthesis that is sufficiently practical for practical use. In addition, 1 or more CS
Since M-synthesis is a synthesis method that does not use a filter, it does not require consideration of stability on the synthesis side. For this reason, mi
, better than the Kti vocoder when transmitting mi information to the synthesis side and reproducing the voice on the synthesis side, when the line quality is relatively poor and errors occur during transmission. One possible feature is that it provides a good sound quality.

(Example) Next, the present invention will be explained in detail using examples.

For convenience of explanation, the present invention will be described using an analytical synthesis system that includes the present invention.

FIG. 4 is a block diagram showing one embodiment of the present invention.

This embodiment consists of a transmitting side 1 and a receiving side 2.

The transmitting side 1 further includes an A/D converter 1o1, a Hamming window processor 102, an autocorrelation coefficient measuring device 103, a CSM analyzer 104, a CSM quantizer 105, and a power correction quantizer 10.
6. Pitch extractor 107, voiced/unvoiced sound determiner 108
and a multiplexer 109.

Further, the receiving side 2 further includes a demultiplexer and decoder 201, an interpolator 202, a voiced sound/unvoiced sound switch 2
03, period calculator 204, random number generator 205.3,
Variable frequency generator with phase reset function 206-1, 20
6-2. ..., 206-fL, variable gain amplifiers 207-1, 207-2.・・・・・・207
- module, addition synthesizer 208, variable length window function generator 209,
It includes a multiplier 210 and a multiplier 211.

Now, the operation of this embodiment is as follows. The audio waveform to be transmitted is sent to the A/D via input line 1000.
The output is supplied to a converter 101, where the amplitude and time axis are converted into quantized digital data, and the outputs are sent to a Hamming window processor 102 and a pitch extractor 10, respectively.
7, is supplied to the input side of the voiced/unvoiced sound determination block 108.

The digital data supplied to the Hamming window processor 102 is subjected to weight multiplication using a known Hamming window function for each predetermined frame, and is supplied to the autocorrelation coefficient measuring device 103 for each frame of data. Ru.

The autocorrelation coefficient measuring device 103 calculates the lowest N autocorrelation coefficients νt (where t=1.2..
... Find N).

In other words, the data for one frame is Xt (where t=
o, 1. ..., M-1), each of N t't is obtained by performing the calculation process.

The thus obtained vtcvm for each frame is supplied to the next CSM analyzer, and is also supplied to the power correction quantizer 106 as power information therein.

Now, the CSM analyzer 104, which has been supplied with the set of autocorrelation coefficients for each frame described above, calculates the intensity of each of the CSM sine waves of the corresponding frame by performing calculations to be described in detail later. m2 to specify the angular frequency. ωL
(where i = 1.2...le),
This is supplied to the CSM quantizer 105.

The CSM quantizer 105 is a direct part constituting the present invention and will be described in detail separately, but its outline is as follows instead.

The CSM quantizer 105 converts the set of values of 1ni and ωt into α=max(ryt, , m2 . . . -5) obtained from the set of amplitudes (m, , m2 . . . -5).・、
m%), and outputs the α as correction data to the power quantizer 106, and uses the a to calculate the normalization coefficient α (door □9m2...1
1) is a means of normalizing and quantizing,
As the number of quantization bits, an appropriate number of bits determined by taking into consideration the requirements for reproduction sound quality and the transmission capacity of the line is selected.

The CSM quantizer 105 quantizes the set of values of -2ω, and then supplies it to the multiplexer 109.

Also, the above-mentioned V. The power quantizer 106 also receives the normalization coefficient α and the normalization coefficient α. After being quantized with an appropriate coarseness determined from the above-mentioned viewpoint, it is similarly supplied to the multiplexer 109.

Further, the multiplexer 109 receives data obtained by appropriately quantizing the digital original audio signal from the A/D converter 101.
Similarly, the voiced/unvoiced sound determiner 108 also determines voiced/unvoiced sound based on the supplied digital data, and supplies this to the multiplexer 109 as a binary signal. The multiplexer 109, which has been supplied with the above signals, combines these signals into a form that can be easily separated on the receiving side and is suitable for transmission over the given transmission path, and then combines the signals into a form that can be easily separated on the receiving side and is suitable for transmission over the given transmission path. It is transmitted to the receiving side ・2 via.

Now, on the receiving side 112, the thus transmitted signals are decoded and separated in the demultiplexer and decoder 201, thereby restoring each signal on the input side of the multiplexer 109 on the transmitting side 1.

Each signal thus restored is supplied to an interpolator 202 having a memory function, and after performing necessary interpolation, it is used as follows.

First, the CSM specifying the acceptable angular frequencies (Sushi~ω7) is added to the frequency control inputs of the variable frequency oscillators with phase reset function 206-1 to 206-2, and these Set the output angular frequency of the oscillator to the specified angular frequency ω, ~ω.

Furthermore, m1 to mn, which are designated as the acceptable strengths (power amplitudes) of the CSM, are the variable gain amplifiers 207 to 1 to
The signal is supplied to the gain control terminal of the 207-channel, thereby controlling the oscillation power of each frequency to a specified value.

The thus obtained number of blades are subjected to countable composition in countable combiner 208, and then supplied to the next multiplier 210.

Now, the pitch period information output from the demultiplexer and decoder 201 is interpolated as necessary in an interpolator 202 including a memory, and is converted into digital data representing the pitch period by a voiced sound/unvoiced sound switch 203.
supplied to

On the other hand, the random numbers generated by the random number generator 205 are supplied to the pulse interval calculator 204, where they are converted so that the random number distribution width and its lower limit value become specific values, and the phase reset time interval during unvoiced sound is calculated. is supplied to the other input of the voiced sound/unvoiced sound switch 203 as a data string for determining the voiced sound/unvoiced sound switch 203.

Further, a binary signal (V/U) that distinguishes between voiced and unvoiced sounds output from the demultiplexer and decoder 201 is supplied as a switching control signal to the above-mentioned switch 203, and in the case of a voiced sound, the switch 203 selects the digital data representing the aforementioned pitch period output from the interpolator 202 and supplies it to the window function generator 209 .

Furthermore, if the binary signal (V/U) specifies an unvoiced sound, the switch 4203 selects the data string side representing a random time interval generated in the stochastic process of the output of the periodic calculator 204 described above. , is supplied to the window function generator-a 209 instead of the above-mentioned example of digital data representing the pitch period.

Now, the window function generator 209 is used to generate a window function that ensures the continuity of the audio waveform by removing discontinuities that occur in the output waveform due to phase reset, and also has a close time relationship with this window function. It also generates a phase reset pulse.

As mentioned above, a data string specifying the interval between the next phase reset pulse is input to the window function generator 209 via the switch 203. impulses having a time interval of Perform phase reset. It is also supplied to the interpolator 202 via line 2090 and used as a timing signal for interpolating the angular frequency data ωt and intensity data mi.

Now, the window function generator 209 generates the following variable length window function W in synchronization with the generation of the above-mentioned phase reset pulse.

That is, let T be the value of the inter-nox interval for phase reset at that point in time specified by the input data,
If the elapsed time since the previous phase reset pulse was generated is X, then W = 0.5 + 0.5 cos (π-) (g) T However, generate a window function expressed as o<a-<'r. do. This window function W(
3:) is shown in FIG. 5 (6). The above-mentioned value of T represents the pitch period in the case of voiced sounds, and represents a variable that occurs in a stochastic process in the case of unvoiced sounds, so it changes over time. Therefore, this window function w(l+) has a variable length and is synchronized with the generation of the phase reset pulse mentioned above in the relative time relationship shown in the fifth example (the start and end points of the window function). (This almost coincides with the time point at which the phase reset pulse is generated.)

The window function thus generated is provided to multiplier 210 via line 2091. As a result, the multiplier 210 generates a sine waveform whose phase is reset for each phase reset pulse synthesized by the addition synthesizer 208, and the above-mentioned window function generated in synchronization with each phase reset pulse. The product with W(,) is obtained. The waveform obtained in this way is expressed by the window function W(,) immediately before the phase of each sine wave is reset.
The waveform is continuously converged to 0 by the multiplication of Continuity can be removed.

The output of the multiplier 210 from which discontinuities have been removed is supplied to the next multiplier 211, where it is weighted by the power information of each frame sent from the transmitter 1, and output as synthesized speech from the line 2000. .

As explained above, on the receiving side 2 of this embodiment, the CSM synthesis necessary for the above-mentioned speech synthesis is executed, and as a result, the reproduction of the original voice input to the transmitting side 1 is
Despite the compression of the amount of information and transmission errors in 20G, the sound quality is relatively good.

The interpolation for each transmission data by the interpolator 202 described above is performed using various combinations (for example, only towns or ω
, -1, etc.), and the interpolation method is also possible using linear interpolation or interpolation using a higher-level function. Regarding interpolation for ωi*mi, it is advantageous to select interpolation points so that interpolation data can be obtained at each point in time when the above-mentioned phase reset pulse is generated, and the values of ωi and mi can be updated at this timing. To do this, a phase reset pulse is provided to interpolator 202 via line 2090 as described above.

In order to perform such interpolation, the interpolated data is obtained after the required subsequent data arrives or is generated, so the phase reset and frequency ω for the oscillator 206 and the setting of the frequency ω, and the intensity for the amplifier 207 are r
Actual processing such as setting ni will be executed after a necessary fixed time delay from real time. Therefore, interpolator 2
02 includes a memory for storing necessary information until a necessary point in time.

Next, a circuit example of the variable frequency oscillator with phase reset function 206 is shown in FIG. A voltage applied to frequency control terminal 2061 causes constant current to flow to power supplies 2062 and 2063. The charging/discharging current value for the capacitor 2064 is controlled, thereby making the oscillation frequency variable. The oscillation voltage waveform at one point is a triangular waveform that linearly rises and falls between +Vr and -Vr of the reference voltage. When an impulse is applied to the phase reset terminal 2065, the υ point is momentarily grounded and forcibly pulled back to 0 potential, from which the oscillation is restarted and phase reset is performed. This one-point triangular wave oscillation output is input to a sine wave converter 2066, converted to a sine wave, outputted from a terminal 2067, and used as the output of the oscillator 206. The sine wave converter 2066 can be easily realized, for example, by reading out a sine function value stored in a ROM as an input waveform.

Further, such a variable frequency oscillator with a phase reset function can be easily realized using a computer program.

Next, a circuit example of the variable gain amplifier 207 is shown in FIG. The signal to be amplified is applied to terminal 2071, and the control signal is applied to terminal 2
072 to control the amount of negative feedback and obtain an output having a controlled amplitude at the output terminal 2073.

In addition, it can also be realized by using an analog multiplier, or by using an analog waveform input as the reference voltage of the D/A converter and using control information expressed in digital quantities as the digital input. It can also be easily realized by a method.

Next, an example of a circuit of the random number generator 205 is shown in FIG. 15
A stage left register 2051 and one middle adder 2052
and K) generate a 15th order M-sequence of pseudo-random numbers with a period of 2 -1. Clock terminal 2053 at the required time
By adding a shift pulse to , the following random value is obtained.

Next, a block diagram of the period calculator 204 is shown in FIG. 9(6). This is the 0 output from the random number generator 205 mentioned above.
This converts random numbers that are uniformly distributed in the range from 2 to 1 to a distribution suitable for use as random numbers that specify the time interval of phase reset pulses during unvoiced speech, and uses a constant multiplier 2041 and a constant addition. It consists of a container 2042. As a result, as shown in FIG. 9(B), the random number distribution width and lower limit value can be set to appropriate values.

Next, an embodiment of the window function generator 209 is shown in FIG.

This includes a register 2091, a presettable down counter 2092, a counter 2093, and a read-only memory (
ROM) 2094.

Data T specifying the phase reset pulse interval supplied from the switch 203 is stored in the register 2091. The down counter 2092 is a high-speed clock C with a constant period.
First, the register 209 is a counter that counts LK.
The content T of 1 is preset and counted down using the clock CLK. The contents of counter 2092 are 0
When this happens, a pulse is generated from the output terminal, which presets the contents of the register 2091 again and starts counting down this value. Thus the down counter 209
A pulse train having a period proportional to T (for example, T/&) is generated at the output 2092-1 of No. 2. This pulse train is added as a clock to the color/receiver 2093.

The count output 2093-1 of the counter 2093, which is incremented by this clock, is added to the ROM 2094 as an address designation signal, and the window function W (,
), are read out in order and output to line 2091. When the contents of the counter 2093 reach k, the ROM 20
The last data of window function W(Y) of line 2094 is read out, and at the same time, the counter 2093 is reset and
A reset pulse is output at 90. This reset pulse is used as the above-mentioned phase reset pulse supplied to the phase reset terminals of the oscillators 206-1 to 206-f& and the interpolator 202, and
It is used to set the next input data to . Further, data of the window function w(6) previously stored as individual samples in the ROM 2094 is output to a line 2091 and supplied to the multiplier 210. In this way, phase reset pulses whose inter-pulse intervals have successively specified values and a variable-length window function W(, ) synchronized with these pulses as shown in FIG. 5(B) are generated.

Next, CSM analysis will be explained.

As mentioned above, C3lvl1 analysis uses, for each analysis frame, N low-order tag values of the sample autocorrelation coefficients directly calculated from the speech waveform to be represented, and a composite wave (rL sine waves). The frequency of each sine wave to be synthesized is adjusted so that the N low-order tap values of the sum of K (power amplitude) are matched. 1Lc.

Now, if the autocorrelation coefficient of tap t of the composite wave is rt, then as mentioned above, rt=, J, 771i CXm l-machi.

On the other hand, from the sample Xt of the audio waveform to be expressed, the sample autocorrelation coefficient ut of tap L of a certain frame is.

From this, rt=vt...(2)t
= 0, 1, 2, --...N However, N=2n
When set to -1, the following matrix expression is obtained.

However, the above equation cannot be solved by simple matrix operations because ω and 7ni are unknown. Therefore, set ω, −Yu X□・Be4) and perform the substitution of 3Tt (”=) . . . −(5) to cm L GJ 6 = cos t cos−1.

This Tz(x) is a Tcbebychef f (Chebychev) polynomial. When this substitution is performed, equation (3) is converted as follows.

However, in general, 1 is To (:I:), Ts
(,2?) ------・Linear combination of T z(x) and 1
It can be represented by -.

That is, where 5(4) is the inverse Tchebycheff coefficient.

The linear combination At is defined as shown below.

However, t=0.1, 2. . . . 12n-1 Then, by using the relationship of the equation (7) and the equation (8) on the left and right sides of the equation (6), the following relational expression is established.

Now, here, xl, x2. ..., define a first-order polynomial with a zero point at x7, use this Pa(3:) to create an equation similar to the left side of equation (9), and examine it. . It is clear that the above equation is 0, but it can be further rewritten as follows.

From the above, t=0, 1, 2. ...The following formula is obtained as n.

However, since 吃゛←), = 1, holds true. The matrix formed by Ai on the left side is generally Hankc
This is called the l (Hangul) matrix. As described above, each person is given by equation (8) from the sample autocorrelation coefficient νj of the speech waveform to be expressed and is known.

Once each p is found, the road equation P, ('') = x''+y? :, -8x"-"+...
・-・As a solution of p0=0, (xl, x2....,
X,) is required.

From this, each CSM frequency ωi is expressed as ωi'' in equation (4)
i, and the CSM intensity mi is determined using the following equation derived from equation (9).

Note that the matrix on the left side of the above equation is generally Vander Mon
This is called a de (7are del monde) matrix.

To summarize the above, the analysis algorithm for CSM analysis is as follows.

(1) Calculate the sample autocorrelation coefficient (2) Define At using the inverse Chebyshev coefficient.

Find the number of 2t.

p, <x) 3-'10F(:jl ” ”-”+ PC:
, 22 "-"+ -+ P(1'= +p 6=0 (5) Perform inverse tuft transform to obtain 08M angular frequency (ωt).

ω = tuft Xl (6) Solve the Van del Monde (777 Del Monde) matrix equation to obtain the CSM intensity (-i).

Each angular frequency (ω1.ω2...ω,) and acceptance strength (ml, m2...mn} can be determined.

Note that, as an efficient method for solving the above-mentioned Hankel matrix equation, a method is known in which initial conditions are given and solutions are sequentially obtained.

Furthermore, since it has been proven that the above-mentioned path-order algebraic equation has only real roots, the roots can be found using the Neaton-Raphson method or the like.

Furthermore, as an efficient method for solving the Vander Monde (77y Del Monde) matrix equation, a method of sequentially obtaining solutions by performing triangular matrix formation can be used. The above analysis method is based on Sagayama et al.'s paper 1: Speech spectrum analysis using a composite sine wave model, "Transactions of the Institute of Electronics and Communication Engineers," 81/2 Mol, J64-A tJn 2p.
, ios~112.

Finally, the CSM quantizer 105 and power correction quantizer 106, which are direct parts constituting the present invention, will be explained in detail using the drawings. FIG. 11 is a block diagram for explaining the CSM quantizer 105 and the power correction quantizer 106 in detail.

The CSM analyzer 104 specifies the intensity and angular frequency of each of the n sine waves of ficsM;
．． 2.・・・・・・Le) pair is temporary memory (1), 10
51. The temporary memory (1) 1051 outputs the mi to a normalization coefficient searcher 1052 and a CSM intensity normalizer 1053. The normalization coefficient search unit 1o52 searches for the normalization coefficient α and the maximum frequency number according to the following procedure.

(1) Set the initial state a cL=m L I I = l.

(2) Investigate the magnitude relationship between α and m2.

If a>mz, execute (4) next.

If a<z2, perform (3) next.

(3) α=m3. Set I=2.

(4) Investigate the magnitude relationship between α and rlL3, and perform the same processing as in (2) above.

(5) Below m4...The same process as in (4) is performed up to shoulder N.

The normalization coefficient searcher 1052 sends the searched α to the 1 upward corrector 1061 and the CSM intensity normalizer 1053, and
The I'ir: output to the CSM intensity quantizer 1054. The CSM intensity normalizer 1053 is a temporary memory (1).

Using the normalization coefficient α, the mi supplied from 1051 is calculated as 1ni=71! i/α (however, i = 1, 2
, -=...rL). Furthermore, the CSM intensity normalizer 1053 calculates the square root of the nine doors V;]-(' = 1
* 2 *...L) Determine the fog and output it to the CBM intensity quantizer 1054.

The CSM intensity quantizer 1054 is the normalization coefficient searcher 105
2 and the numbering machine of the maximum frequency supplied from 2 and the r (L
ml, 2. For example, the first
Linear quantization is performed using the bit allocation shown in FIG. 2, and the quantized data is output to the temporary memory (2), 1056.

Next, the format of quantization will be explained with reference to FIG.

Figure 12 (a) is the 9th order CSM analysis (corresponds to Le÷5)
The resulting CSM intensity is 77L,...
This figure shows the pit distribution for quantizing the door with 16 bits. The highest CSM strength is specified in correspondence with the numbered work. Here, when the strongest CSM has a number I of 1, 10# is given to the bit shown at the left end in the figure as shown in FIG. 12 0) to α, and I is 2.3, 4.5
In this case, as shown in FIG. 12(1)-be, 61'' is given to the n bit shown at the left end in the figure.

By the way, when we investigate the distribution of CSM strength, it is extremely common for myu to have the strongest CSM strength. FIG. 13 shows the CSM intensity yx1,2+1. , ....-1 is normalized by calculating the normalization coefficient α, and in the figure, the number written as "@7 frame number" is the number of frames where the intensity is the strongest. The total number of analysis frames is 6963. That is, the ratio of ㅝ with the strongest CSM strength is 5895/69.
63 and 0.847, and as shown in FIG. 12(b), the configuration is such that ml is the strongest (when I=13, the designation of I can be performed with the least number of bits).

Note that the strongest CSM strength itself is normalized by itself, so it is always 1. o, and no information transmission is required.

Returning to FIG. 11 again, the thus quantized CSM intensity parameters are output to temporary memory (2), 1056. The CSM frequency quantizer 1o55 specifies the angular frequency of the CSM roadside sine wave ω1 (however, ' 1 *
2 +”−−−・n) +7)!11t− Temporary memory (
1), which receives the supply from 1051, performs linear quantization taking into consideration the distribution range of each ω, which has been investigated in advance, and outputs the quantized data to temporary memory (2), 1056. Temporary memo IJ (2), 1056 is i quantization CSM strong 1
i and CEiM angular frequency data to i Lunaplexer 109
Output to. The power corrector 1061 receives the power from the autocorrelation coefficient measuring device 103, multiplies the nine-dimensional data by the coefficient α supplied from the normalization coefficient searcher, and then applies the result to the power quantizer 106.
Output to 2. Power quantizer 1062 converts the result to 17
After converting to squared amplitude information, for example dμ2ssPCM
The non-linear quantization used in the above is performed and output to the multiplexer 109.

Note that the denormalization on the synthesis side is automatically performed by the multiplier 211.

(Effects of the Invention) As described above, the present invention has the effect of increasing the efficiency of quantization by quantizing the CSM intensity parameters in consideration of the mutual relationship between the CSM intensity parameters.

[Brief explanation of drawings]

Fig. 1 is a diagram showing an example of the speech feature vector Batano using CSM parameters, Fig. 2 is a diagram showing an example of the correspondence between the CSM line spectrum and the LPG spectrum envelope obtained from the same speech sample, and Fig. 3 (A) is a diffused CAM
Figure 3 showing the spectral envelope and pitch fine structure of
Figure (B) is a diagram showing a CSM spectrum obtained by simple addition. Figure 4 is a block diagram showing an example of an analysis and synthesis system including the present invention. Figure 5 (A) is a functional form of a variable length window function. 5(B) is a diagram showing the relative time relationship between the variable length window function and the phase reset pulse. FIG. 6 is a diagram showing an example of a circuit of a variable frequency oscillator with a phase reset function. Fig. 7 is a diagram showing an example of a circuit of a variable gain amplifier, Fig. 8 is a diagram showing an example of a circuit of a random number generator, Fig. 9 (A) is a block diagram of a period calculator, and Fig. 9 (B) is a diagram showing an example of a circuit of a random number generator. FIG. 10 is a block diagram showing an example of a variable length window generator, and FIG. 11 is a diagram showing the distribution of random numbers output from the period calculator, and FIG. Block diagram, 1st
Figure 2 shows an example of quantization format, Figure 13 shows CAM
FIG. 3 is a diagram showing an example of intensity distribution. In the figure, 1... transmitting side, 2... receiving side, 101... A/D converter, 102...
... Hamming window processor, 103 ... Autocorrelation coefficient measuring device, 104 ... CSM analyzer, 105
...... CSM quantizer, 106... Power quantizer, 107... Pitch extractor, 108 Voiced/unvoiced sound determiner, 109... Multiplexer , 201... Demultiplexer and decoder, 202... Interpolator, 203... Voiced sound/unvoiced sound switcher, 204... Period calculator,
205...Random number generator, 206-1 to 206-
3...Variable frequency oscillator with phase reset function,
207-1 to 207-3...variable gain amplifier,
208... Addition synthesizer, 209... Variable length window function generator, 210, 211... Multiplier, 1051... Temporary memory l), 1052 ...
...Normalization coefficient searcher, 1053...CSM
Intensity normalizer, 1054...CSM intensity quantizer, 1055...CSM frequency quantizer, 10
56...Temporary memo! j(2), 1061...
Power corrector, 1062... Power quantizer. Brush stroke [;'' Rate 2 PTA number 0 hour □ Early S figure 6 figure 7 Zu 8th q figure (c) Bō 120 (α) っ l 0 (ryn+ rns
rn4 m l=2 of 4b) N: 7 fil rnz m4
rn 1=3n4 now J, tto ynr
171. rn rn I=4
v>40>e /// ynr 1712
mJ m4
No. 91 No. 2, Title of the invention: 08M parameter quantization method and device 3, Relationship with the person making the amendment case Applicant: No. 33-1 (423), Shiba Go-cho 1, Minato-ku, Tokyo Representative from NEC Corporation Person: Tadahiro Sekimoto 4, Agent 6, Drawing L subject to amendment Contents of amendment Replace Figure 13 with attached Figure 13. 7v secretion /ηtt, r employment y
υYo Iyo - Saitoichi - Kotosanichi - Sanyuichichi χ wo Ganshobi XMI Yumi Sunaichikonichi - Yanuke -

Claims (2)

[Claims] (1) In CSM parameter quantization, which expresses the spectral envelope of speech as a set of sine waves with a predetermined number n of free frequencies and amplitudes, a set of amplitudes {m_1, m_2, ..., m
_n} by a normalization coefficient α expressed as a=max{m_1, m_2, . . . , m_n}. (2) In a CSM parameter quantization device for expressing a CSM parameter in a small number of bits, which expresses the spectral envelope of speech by a set of a predetermined number n of sine waves with free frequencies and amplitudes, the sine wave set of amplitudes {m_1, m
_2,..., m_n}, the maximum value of the set a=max{
m_1, m_2,..., m_n); and means for normalizing the set of amplitudes by the maximum value a found by the means.