JPS607500A - Multipulse type vocoder - 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 relates to a multi-pulse vocoder. Analyze the input audio signal and determine whether audio 1, ? News o7
The spectral envelope information and sound source information that are formed by the analysis side are extracted by the analysis side, and these audio information are sent to the synthesis side via all six transmission paths to generate a human voice signal sound.
[It's been 1.

The above-mentioned spectral envelope information is information on the spectral distribution of the vocal tract system that generates the input speech signal.
The number of L corresponding to the analysis order obtained by PC analysis
The spectral distribution information is removed from the input audio signal in one sentence, for example, the α parameter, and the sound source information indicates the fine structure of spectral discoloration.

This is known as the so-called residual signal, and it contains information regarding the source strength, pitch period, and voiced/unvoiced input audio signal.1This information is usually calculated using the autocorrelation coefficient for each analysis frame of the input audio signal. It is also well known that it can be extracted through

Now, when the input audio signal is synthesized on the synthesis side of the vocoder, the spectral envelope information is usually used as the coefficients of the LPC synthesizer that uses an all-pole chamber digital filter to form an approximate complete vocal tract system, and the sound source information is It is used as a driving sound source for this digital filter, and input audio signals are synthesized by this digital filter.

The conventional LPC vocoder thus obtained is approximately 4
Although it is possible to synthesize speech even at a low pit rate of Kb (kilobits) or less and is often used, it has the drawback that high-quality speech synthesis is difficult even at a high bit rate. The reason for this is that when modeling sound source information, for voiced sounds, the pitch period corresponding to the content is extracted and approximately represented by a single impulse train corresponding to this pitch period, while unvoiced sounds with random periods are is assumed to be a simple modeling process in which it is approximated by white noise, so the sound source information of the input audio signal is not faithfully extracted, and therefore the input audio signal contained in the sound source information is This is because analysis 9 synthesis of waveform information has not been performed.

A multi-pulse vocoder has recently become well known as a type of vocoder that transmits waveforms and synthesizes input audio signals in order to improve the problem caused by non-transmission of waveforms. .

2=fNIJ is a block diagram showing the basic configuration of a conventional multi-pulse reduction vocoder.

LPC @ generator 1 divider 1 is equipped with an all-pole digital filter that simulates the whole, whose coefficient t is the input rr'A child 200
1 input audio signal x(nl(n=1
．． 2.3...n) All LPC analyzers 2 supply LPC coefficients analyzed for each analysis frame. Sound source pulse generation ￥::r 3 is a drive sound source sequence V(nl't: obtained, this'(i)L
It is supplied as a driving sound source for the PC synthesizer 1.

The LPC synthesizer 1 uses the input LPG coefficient as a coefficient of a synthesis filter that normally makes full use of an all-pole digital filter, and outputs a synthesized signal Y(n) driven by a multi-pulse as a driving sound source. In this case, the multi-pulse contains waveform information of the input audio signal, and the PC@-synthesizer 1 synthesizes the input audio signal including the waveform information.

Now, the @-composed signal T(n) output from the LPG synthesizer 1
Next, the subtracter 4 takes the difference from the input audio signal x(n),
The error e(n) is obtained and sent to the perceptual weighting device 5.

The auditory weighting device 5 calculates the following (1) for the error e(n).
The weighting having the characteristic W■) shown in the equation is performed by applying perceptual weighting using a filter and then sending these weights to the squared error minimizer 6.

...... (1) In equation (1), a is the LPC coefficient that should be the coefficient of the all-pole digital filter of the LPC synthesizer 1, and p is its order. Therefore, the LPC analysis order λQ , rij, weighting is a coefficient,
Z indicates Z = exp (jλ) in the transfer function 11 (Z"") according to the Z transformation representation of the all-pole digital filter, λ = 2πΔTf, and ΔT is the sampling period of the analysis frame, f[indicates the frequency .

Also, in equation (1), the weighting is done by the coefficient rf, 0<r<1
It is set within the range of .

+1+2 -jW(Z)it r = 1 two pairs td, 1, for r= 0 W(Z) = t
−p (Z).

The value of 1 is set within the above-mentioned range so as to suppress the excessive level appearing in the formant region of the frequency spectrum of the error e(n).

Auditory weighting role of signals to be synthesized ψ111'ff:
The optimum value is usually selected in advance through an optimum hearing test.

The error e (nl) weighted in this way is the driving sound source series V(n) output from the sound source pulse generator 3,
That is, in order to determine the optimal time position and amplitude of the multi-pulse, it is sent to the square error minimizer 6, which calculates the square error ε according to the 9th order equation (2), and then drives the F to minimize ε.
il+sound source series V(n) is selected.

ε=Σ(e(n)-X-ω(rl)" +++++++...
+ (2) 1?1 In the formula (2), the symbol → (• indicates the auditory weighting is the convolution integral by the filter, and N indicates the interval length for calculating the multipulse.

The above process is repeated for each multi-pulse.
Synthesis by analysis is performed for each multipulse. This is a so-called Analysis-by-8yntbesis method (hereinafter abbreviated as A-b-8 method).

This A-b-8 method is also clear from the above-mentioned inherent nature.Pulse generation for each multi-pulse, square error calculation, and pulse position/loss 11'7! Since this is performed in a loop of N4, although it is an effective means in a low pit rate region, the amount of calculations required is extremely large.

Regarding this A-b-8 method, B, S.

Atal et al., “A New Model
of LPCExc+-tation for Pro
ducing Natural-8oundingSp
eech at Low flit Rates'', P
roc, ICASSP82, TRp614-617. (
1982) and others.

To address these shortcomings in the conventional A-b-8 method, correlation calculation 1. [Based on γ]'1) The following processing algorithm has been recently introduced to efficiently calculate suitable multi-pulses.

That is, the input audio signal x(n) is divided into N lean pulls by processing arrays, and multipulses are comprehensively counted for each frame.

(0, it is assumed that there are several sound source pulses in one analysis frame, and the i-th pulse is at a time position m from the frame draw.)
i, and its amplitude is gi, then LPG
The driving sound source d(b)t of the synthesis filter is expressed by the following equation (3).

d(n) ==, Σfi−δn, ml ・・−group・
(3)+-1 In equation (3), δn and mi are Kronecker's delta functions, and δH,mi = l (n = jil, )
, δn.

mI=O(n~mj).

The LPG synthesis filter is driven by this driving sound source d inl and outputs a synthesized signal.

As an LPC synthesis filter, let us consider, for example, an all-pole digital filter, and if it is expressed by its transmission + JJ number total impulse response k(n) (0≦n≦M-1), one synthesized signal x(n) is It is expressed by the following equation (4).

? (nl = Σd(gJ-h(n--g) ・-・-・
...-t41-0 In equation (4), the d test) penetrates the i-motion sound source.

Next, weighting that performs auditory correction on the error between the input audio signal x (n) and the composite signal ma (n) reduces the error to ew (n
), then e and (n) are expressed by the following equation (5).

%! ”) = (x In) − T I, n)) + v
v:n) --(5) Further, the square error can be derived from the equation (5) and expressed by the following equation (6).

(6) In equation (6), M represents a reasonable number of intervals that minimizes the error, and is selected to be one analysis frame length, for example.

The optimal sound source pulse train is (6) 2k.
Ja is reduced by obtaining Pi'fK: This %i is derived from the above-mentioned equations (3), t4+, and (6) as shown in the following equation (7).

(n-rnz)・hw(n-ryg))/Σh,y(
nm, ).

1-1 h, (n-ml) =...” (7) In equation (7), Xw(n) is x(rl-Xw(n), hw(nli,
J: h (nl % w(n) f is shown. The first term of the numerator on the right side of equation (7) is the mutual correlation function 9'bx (m+) of the time delay mi between nl and h v (n) ), and the function of the second term?), ), (mz m,
) (1≦mtm,≦M)ffi.

The covariance function ψhh (mtm, ) is the autocorrelation function Rb
h< l mt-mt I >, so (
Equation 7) can be expressed as the following equation (8).

(8) According to equation (8), if all pulses are generated at the 9-hour position m1, the amplitude Pi (mi) will be determined as the optimum one. Note that in equation (8), 1≦mi≦M.

After focusing on the sound source pulse and calculating its amplitude using equation (8) at two different time positions, the one that maximizes the absolute value of the amplitude is the one that minimizes the squared error shown in equation (6). By repeating this procedure, a plurality of sound source pulses can be obtained.

Regarding the calculation algorithm mentioned above.

Obama, Araseki, and Ono, ``Study of multi-pulse driven speech coding method'', detailed at the Institute of Electronics and Communication Engineers' Communication Systems Study Group, March 1983.

By generating real multipulses based on such a calculation algorithm, it is possible to calculate an appropriate amount of multipulses from the cross-correlation number, autocorrelation function, and maximum value calculation, so the configuration is very simple. It is possible to realize a multi-pulse vocoder which is subjected to I'+1 elements and which can significantly reduce the calculation time.

However, even with the multi-pulse vocoder improved in this way, there are still problems as described below.

That is, the pulses (, ( than 7
1/- When the number of pulses existing in the signal is large (((is the synthesized signal V), the wave transmission based on the sound source information of the input audio signal No. 13 is not faithfully executed, and the audio product of the synthesized signal No. 9 There will be a degree of deterioration in which T corresponds to the difference in the number of pulses mentioned above.

In a multi-pulse vocoder, for example, analysis frequency JIJ]
The number of sound source drive pulses to be generated in one frame with fc20mSEC is usually 4 to 4 depending on the focus rate.
A preset fixed number out of 16 (2 to 2) is used. When the input audio signal is a high-pitched voice with a small pitch period, such as a female voice or a child's voice.

The pitch period of the sound source signal is 2. It is not uncommon for s m S EC to be 6 degrees, and in this case, at least 8 driving sound source pulses are required to be set in one analysis frame. In such a case, the number of driving sound source pulses to be generated within 9 analysis frames is 8 or less.

For example, when the number is set to four, in a multi-pulse vocoder that utilizes such drive sound source pulses, a synthesized sound containing a result similar to a double pitch error will be generated, and the synthesized sound quality will be significantly degraded.

The purpose of the present invention is to eliminate all of the above-mentioned drawbacks, and to improve the pitch period extracted from a human voice signal when the number of pulses present in an analysis frame is larger than the drive sound source pulses that should be generated within 9 analysis frames in a multi-pulse vocoder. 1T (1 single groove 57 multi-pulse type vocoder) which greatly improves the deterioration of synthesized sound quality by providing a means to use it in place of all multi-pulse i-motion sound source pulses generated by modeling based on It's about doing.

A specific example of the multi-pulse vocoder of the present invention is as follows.

Analyze all of the input audio signals, analyze the entire spectrum of the LPG coefficients extracted by T, PC, and analyze the entire spectral envelope information of the input audio signal, and analyze the sound source information constituting all of the audio information of the input audio signal for each frame. In a multipulse vocoder that analyzes and synthesizes the input audio signal by expressing it with a plurality of impulse sequences (multipulses) having generation time positions jIl and amplitudes corresponding to the characteristics of the sound source information, The number of pulses extracted from the pitch period Kanakura and recorded sound source information existing in the analysis frame for each analysis frame of one speech word is 11.
:lr l, and if this is greater than the number of said multipulses to be generated within the analysis 7 frames, then all said multipulses are replaced by a plurality of impulse sequences corresponding to said pitch group 1. All of the means for generating this information are constructed separately.

Next, ask for the drawing! 1, the present invention will be explained in detail.

FIG. 2 is a block diagram showing an embodiment of the analysis side of the multi-pulse vocoder according to the present invention, and FIG. 3 is a block diagram showing an embodiment of the synthesis side of the multi-pulse vocoder according to the present invention.

The analysis side of the multi-pulse vocoder according to the present invention shown in FIG. 2 includes an LPC analyzer 7. Cross-correlation calculator 8. Pitch extractor9. First encoder 10° autocorrelation function calculator 1
1. Source pulse generator 12. Second encoder 13. Third
The encoder 14 and the multi-blephr 15t are constructed.

The input audio signal input via the input terminal 7001 is
PC analyzer7. The signals are supplied to a cross-correlation function calculator 8 and a pitch extractor 9, respectively.

The LPG analyzer 7 analyzes every frame of the input audio signal,
Quantize as digital]-,''C with a preset number of bits, analyze this quantized audio signal @Lpc, extract the p-order parameter (partial autocorrelation coefficient) as an LPC coefficient, and output it. is supplied to the first encoder 10 via line 701. In this example, the analysis frame V-frame is
It is set to 0m8Ec.

The first encoder lO j of the input L I) C coefficients
I', after all childization and encoding, output In1
001 to the multiplexer 15.

LPC analyzer 7F again. The impulse response h(n) (0≦n≦M-1) is fully calculated from the LPC coefficients, and the output signal 702. 1st encoding & StO, mutual 41 old ↓ I4 cabinet number calculator 8 and self 40 via output lifecar line
is supplied to the interval function calculator 11.

The cross-correlation function calculator 8 fully calculates the cross-correlation function ψh8 by fully utilizing the input audio signal impulse response h(n),
This is Kanayama Karain 801, all valves, sound source pulse generator 1
Send to 2.

In addition, the self-phase four gates 5' (CRhh?-
Tears J.

This is output to the sound source pulse calculator 12 via the output line 1101ffi.

Sound source pulse calculator 12t, input the mutual phase open function ψ,
h and performs the calculation of equation (8) to obtain a predetermined sound source pulse train, and sends the amplitude and position information of these pulses to the second encoder 13 via the output line 1201. , which performs quantization and encoding, and then sends it to the multiplexer 15 via an output line 1301.

In this way, the LPG coefficients and multipulse data that are quantized and encoded and sent to the multiplexer 15 are time-divided in a predetermined manner via the multiplexer 15 as data representing all the spectral envelope and sound source information of the input audio signal. It is.

31 from the analysis side shown in FIG.
The number of multipulses that should be generated within an analysis frame in the processing on the analysis side is set as a fixed number in advance according to the transmission bit rate. High pitch Shu J like a voice! If the number of pulses of the sound source information existing in one analysis frame is larger than the number of pulses of the sound source information that exists in one analysis frame, it becomes impossible to faithfully analyze all of the sound source information and transmit the waveform. As mentioned above, the quality of the audio reproduced on the synthesis side deteriorates in response to the difference in the number of these two pulses, as mentioned above.

Therefore, in this embodiment, the pitch extractor 9 and the third encoder 14 shown in FIG.
Isometal Provision 9 attempts to eliminate this drawback as follows.

When the pitch extractor 9 receives the input audio signal, it calculates the entire autocorrelation function for each analysis frame, and determines the pitch period based on this. This means that the input audio signal is around 44. Q
If the input audio signal has a pitch family of 4jjl and the delay time is the same as the pitch family of be.

The pitch extractor 9 extracts the pitch range m* for each analysis frame of the input audio signal, and this pitch family 1iJJ is the second pitch family 1iJJ.
In the multi-pulse type vocoder shown in 1a, when it is determined that the number of multi-pulses is greater than the fixed number of multi-pulses set in advance, W from the analysis 101 shown in ff1i21:4 is performed.
3 Synthesis shown in L\1 (transmit to ill) to generate a multi-pulse modeled based on the detected pinch period instead of a preset number of multi-pulses, and use this as sound source information, that is, a vomiting sound source. LPG synthesis is performed as a pulse, and the rlb production is performed as follows.

That is, the pitch extractor 9 extracts all pitch periods for each analysis frame, compares them with a pitch period judgment threshold value set in advance and corresponding to the number of multi-pulses, and selects a pitch period period smaller than this judgment threshold value. Pitch 4 of analysis frame with? Output line 901 ? temporal position information and amplitude information of the first and last pulses of the sound source pulses existing in the analysis frame of f-report 9 or only the first pulse. It notifies the synthesis side that the number of excitation pulses in the eight analysis frames is greater than the number of multipulses. A multi-pulse alternative control signal with a binary logic value (IN) level is outputted via an output line 902 and sent to all multiplexers 15.

The third encoder 14 aggregates and encodes all the data related to the various information inputted from the pitch extractor 9, and sends it to the multiplexer 15 via these output lines 1401'lr.

Multiplexer 15 thus connects pinch extractor 9
If the pitch period jiQ extracted by I, P'C coefficient data received via the output rhino 1001 are equal to or less than the number of :, 1+ units.

The transmission sound of the multi-pulse data received via the output line 1301 is simultaneously transmitted in a predetermined time-sharing manner via the output line 15 (11K), and the pitch 1 period extracted by the pitch extractor 9 is is smaller than the pinch period of the multi-pulse, i.e. the number of pulses of the multi-pulse to be generated at V-me is greater than the number of source pulses present in the analysis frame. In addition, the above-mentioned data received through the output line 1/1401 and the output line 902
Metal block CL/-(□-1 digit 6 Multi-pulse alternative Wi control signal is included and transmitted simultaneously in time division.

The synthesis side shown in FIG. 3 synthesizes input audio signals based on data transmitted from the synthesis side via a transmission path 15o1, and a demultiplexer 16. First decoder 17. Second decoder 18, third decoder 19.
Switcher 20. LPC synthesizer 21. LPF (Low Pa
ssF'lter) 22 and an alternative sound source pulse generator 23, etc., are installed.

The demultiplexer 16 restores the various data input via the transmission line 1501f to the state before conversion by the time division transmission format of the multiplexer 15, and the LPC coefficient data is sent via the output line 161 to the decoder 17 of ε1↓l. Then, the multi-pulse data is sent to the second decoder 18 via the output line 162, and the pitch period of the analysis frame in which all pitch periods below the judgment threshold are extracted, the first and last pulses of the sound source pulses, or the first pulse. The temporal position information and its amplitude information are respectively supplied to the third decoders 19 via output lines 163, and these decoders perform all the decoding of the data. picture.

They are sent to output lines 171, 181, and 191, respectively.

Also, Demultigrex 16'! f: The multi-pulsing/drilling control signal data output via the output line 16
4 via the switch 20 and the alternative sound source pulse generator 23
will be sent to.

The relay changer 20 receives a multi-pulse alternative input via the output line 164 (when there is no I11 signal, the pip-extractor 9 of the analyzer f111
, the pitch period of the analysis frame is yp equal to or longer than the pitch period of the multipulse, and the multipulse generated within nine analysis frames is the reason why the pitch period of the sound source pulse present in the analysis claim is ) or more than 9 multi-pulses should be generated even if used as a sound source stop, the synthesized sound [J The signal is switched to be sent to the LPG synthesizer 21 via the ffi.

The LPC synthesizer 21 uses the thus inputted multipulses as sound source information for the driving sound source of the p-order all-pole digital filter, and also uses the p-order LPG coefficient data input via the output line 171t as the sound source information. This LPG composition filter is controlled as a link to the all-pole type digital filter to synthesize the input audio signal, and sends it to I and PF 22 via the output line 211, and performs all the predetermined low-pass filtering to convert it into an analog signal. Output line 221 as synthesized voice of
Send to.

Now, when the multi-pulse alternative control signal of logical value II is supplied to the switch 20 via the output line 164, that is, the pinch period of the analysis frame extracted by the pitch extractor 9 on the analysis side is equal to or less than the decision threshold value. The size is also small,
Therefore, due to the multipulses generated within the analysis frame, when the number of source pulses present in the analysis frame is large.

The switch 20 receives an alternative sound source pulse from the alternative sound source pulse generator 23 through the Jinshan line 231 and outputs it as an alternative sound source pulse in place of the sound source pulse data sent out through the output line 201. 202
IJI: is switched to be supplied to the LP'C synthesizer 21 via the LP'C synthesizer 21.

The alternative sound source pulse generator 23 sends data on the pitch period of the analysis frame whose pitch period is smaller than the decision threshold value via an output line 164 and also to an output line 191'.
The time fL position of the first pulse and the last pulse of the sound source pulses included in this analysis frame is determined through 11 and 4.
Enter all the data related to the tumor width information or the time and frequency information of only the first pulse.

The spectral envelope information and the sound source information for the analysis frame are sampled directly. The change is gradual and can be regarded as a stationary signal.Therefore, the pitch period of the sound source pulse existing in the analysis frame for one period stI]t is compared to its magnitude. It is modeled using the pitch period extracted by the pitch extractor 9 based on the temporal position and thread width information of the first and last pulses of the sound source pulses. Impulse sequences of PILl can be easily generated, and all of the time position information of the first and last pulses of the sound source pulses can be used, so that all of these multiple impulse sequences are included in the analysis frame. I can buy it.

The alternative sound source pulse generator 23 outputs a multi-pulse group generated on such an internal one-stroke impulse sequence and modeled with this analysis pitch period as an alternative multi-pulse to the normal multi-pulse, that is, as an alternative sound source pulse. 231.

Note that if the sound source pulse to be used in generating this alternative sound source pulse is modeled as only its first pulse, all nine generated alternative sound source pulses will not be included in the analysis frame, and the last pulse of one alternative sound source pulse will not be included in the analysis frame. exists in common with the following analysis frame. The only difference is that the existence of so-called fractional pulses causes a corresponding deterioration in the reproduced sound quality.

The alternative sound source pulse generated in this way is transferred to the switch 20
The LI'C synthesizer 21 VC is sent to the LI'C synthesizer 21 VC instead of the normal multi-pulse, and the LPC analyzer 21 uses this substitute multi-pulse as a driving sound source and LPc filter filter 17 to perform LPC synthesis for reproducing the input audio signal. Analysis 1 synthesis of input audio signals can be performed without deterioration of sound quality fci=< even when there are many sound source pulses present in an analysis frame. can.

Nao, 21 (2L, <Ikp j and F43 B::,I
K indicates misfire'jl (7) '4% 7#[11;'
Although the I parameter is used as the PC coefficient, other LPC coefficients such as the α parameter may also be used.
It is clear that the same structure can also be implemented as a four-component structure, in which the decoder and demultiplexer are integrated.

Also, if we put LPC synthesis filter t as a non-polar type digital filter other than all-pole type, even if 1 is used, the actual IN is almost the same as t.
Shiuroko is also obvious.

As explained above, according to the present invention, in a multipulse vocoder, when the number of sound source pulses existing in an analysis frame is greater than the number of multipulses that should be generated within one analysis frame, the pulses extracted from the analysis frame are Did you model it after Totomo? Analysis and synthesis of the input audio signal is performed using a means for using the impulse sequence of J[ as an alternative sound source pulse for replacing the multi-pulse, so that the input audio signal has a short pitch period such as a female voice or a child's voice. This has the effect of realizing a multi-pulse vocoder that can significantly improve the quality of reproduced sound even in the case of high-pitched voices.

[Brief explanation of drawings]

Fig. 11.1 is a block diagram showing the basic structure of a conventional multi-pulse vocoder, Fig. 2 is a block diagram showing an embodiment of the analysis side of the multi-pulse vocoder according to the present invention, and Fig. 3 is a block diagram showing the basic structure of a conventional multi-pulse vocoder. FIG. 2 is a block diagram showing an embodiment of the synthesis side of a multi-pulse vocoder according to the present invention. 1...T, P C@-Bouki, 2...
I, PC analyzer, 3...Sound source pulse generator, 4
・・・・・・Reducing W device, old and new...Sensitive weighting is device, 6...
...square difference minimizer, LPG analyzer for 7...,
8... Mutual phase function calculator, 9... Pitch extraction i: circle 1o... Encoder of broom 1, 1
2... Sound source pulse generator, 13... Second encoder. 14... Dan Houki 3 encoder, 15... Multiplexer, 16... Demultiplexer, 17...
...First decoder, 18...River@2 decoder, 19...°...Third decoder, 20...
・・・Chi U changer, 21... LPC synthesizer, 2
2...LPF, 23...Alternative Takehara pulse generator. Agent: Patent Attorney Uchihara Uchihara

Claims (2)

[Claims] (1) LPC (L
Inear Prediction Coeffici
ent, linear prediction coefficient) analyzed and extracted LPC coefficient ik
The spectral envelope information is used together with the spectral envelope information to analyze the sound source information forming the audio information sound 4 of the input audio signal. In a multi-nox vocoder that analyzes and synthesizes the input audio signal by analyzing and synthesizing the input audio signal by generating a plurality of predetermined innolux sequences (multipulses) having a whisper, the input audio signal is analyzed. A means for extracting a pitch period for each frame, using a quotient value obtained by multiplying the analysis frame length by 7111 times the extracted pitch period as a basis, and generating a number of impulse sequences corresponding to this quotient value. A multi-pulse vocoder characterized by: (2) A multi-pulse vocoder characterized in that the means is means for generating all impulse sequences corresponding to the quotient value when the quotient value is larger than the plurality of predetermined numbers.