JPS60239799A - Multipulse type vocoder - Google Patents

Abstract

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

Description

TECHNICAL FIELD The present invention relates to a multi-pulse vocoder, and more particularly to a method for determining the amplitude and position of multi-pulses on the analysis side.

(Traditional technology) Analyzes the input audio signal, extracts spectral envelope information and sound source information that make up the audio information of this input audio signal on the analysis side, and sends this audio information to the synthesis side via a transmission path. Vocoders that reproduce input audio signals are well known.

The above-mentioned spectral envelope information represents the spectral distribution information of the vocal tract system that generates the input speech ig, and usually includes parameters in the number of LPG coefficients, such as the α parameter, corresponding to the analysis order obtained by LPC analysis. The sound source information indicates the fine structure of the spectral envelope, and is known as the so-called residual signal, which is obtained by removing the spectral distribution information from the input audio signal. and voiced/unvoiced information.It is also well known that this information is usually extracted via an autocorrelation coefficient for each analysis frame of the input audio signal.

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

Conventional LPC ho obtained in this way: 7-p
- is widely used because it is capable of 5T speech synthesis even at a low bit rate of about 4 Kb (kilowatt) μ, but 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, the pitch period corresponding to the content of a voiced sound is extracted and approximately represented by a single impulse train corresponding to this pitch period, and the unvoiced sound with a rankum period is is based on a simple modeling process in which the sound source information of the input audio signal is approximately expressed using white noise, so the source information of the input audio signal is not faithfully extracted. This is because analysis and synthesis of waveform information has not been performed.

A multi-pulse vocoder is recently well known as one of the vocoders that synthesizes input audio signals by transmitting waveforms in order to improve the problem caused by non-transmission of waveforms.

FIG. 1 is a block diagram showing the basic configuration of the analysis side of a conventional multi-pulse vocoder.

The LPG synthesizer 1 is equipped with an all-pole digital filter that simulates the vocal tract, and its coefficients are determined by the input audio signal x(n) (n=1.2.3) input via the input terminal 2001.
.

The sound source pulse generator 3 obtains a driving sound source sequence V (n) consisting of a plurality of impulse sequences, that is, multi-pulses, from the sound source information of the input audio signal, and outputs the driving sound source sequence V (n) to the LPC synthesizer 1.
supplied as a driving sound source.

The LPC synthesizer 1 uses the input LPC coefficients as coefficients of an association filter that normally uses an all-pole digital filter, and generates a synthesized signal driven by a multi-pulse as a driving sound source. Output (n). In this case, the multi-pulse contains waveform information of the input audio signal, and the LPC synthesizer 1
synthesizes the input audio signal including waveform information.

Now, the synthesized signal 1(n) output from the LPG synthesizer 1 is then subtracted from the input audio signal x(n) by a subtracter 4,
The error e(n) is obtained and sent to the perceptual weighting device 5.

The auditory weighting unit 5 calculates the following (1
) The weighting having the characteristic w(z) shown in equation (2) is performed by applying perceptual weighting using a filter and then sending these weights to the minimum square error specific volume 6.

] ・・・・・・・・・・・・・・・(1) In equation (1), ak is the LPC coefficient that should be the coefficient of the all-pole digital filter of LPC synthesizer 1, and p is its order Therefore, the LPC analysis order, r is the weighting coefficient, and 2 is z=exp(jλ) in the transfer function H(z-'') by the two-transform representation of the all-pole digital filter, where λ=2πΔT
f, ΔT is the sampling period of the analysis frame, and f is the frequency.

Further, in equation (1), the weighting coefficient r is set in the range O<r<1.

w(z) shown in equation (1) is 1 for r=1, and r=O
for w (z) = 1-1) (Z), and the value of r corresponds to the extent to which excessive levels appearing in the formant region of the frequency spectrum of the error e (n) are suppressed. It is set within the range mentioned above, and plays the role of aural contribution of the signals to be synthesized, and its optimum value is usually selected in advance by an optimum auditory test.

The error e (n) weighted in this way is the driving sound source series V (n) output from the sound source pulse generator 3.
, i.e., the RJM time position and amplitude of the multi-pulse are sent to the squared error minimum ratio 6, and the following (2
), and the drive sound source 7 table V (n) is selected so as to minimize ε.

ε= Σ(e (n) 4w (n) 〕2−−・−=
C2) n=1 In the equation (2), the symbol I indicates the auditory weighting unit 5's convolution integral by the filter, and N indicates the interval length for calculating the multipulse.

The above six principles are repeated for each multi-pulse,
This A-b-8 method is called analysis-by-8yntesis (hereinafter abbreviated as A-b-8 method) in which synthesis by analysis is performed for each multipulse.
As is clear from the above, the b-8 method is performed in a loop of pulse generation, square error calculation, and pulse position/amplitude adjustment for one multipulse.
Although it is an effective method in the FF4M range, it has the disadvantage that the amount of calculation required is extremely large.

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

Atal et at, “A New Model”
of LPCExcit-ation for Pro
ducing Natural-8ounding 5
peechat Low Bit Rates”, P
roc, ICA38F 82°1) p 614-61
7. (1982) and others.

To address this shortcoming of the more conventional A-b-8 method, the following operation 11. Processing algorithms have been recently introduced.

That is, is the input audio signal x (n) N? Every 7 pulls are also divided into logical frames, and multipulses are comprehensively calculated for each frame.

Now, it is assumed that there are several stone source pulses in one analysis frame, and the first pulse is at a time position m from the frame end.
and its amplitude is gl, the driving sound source d(n) of the LPC synthesis filter is expressed by the following equation (3).

d(n) = Σg+・δn, ml ・・・・・・・・・
...(3) Plate=1 In equation (3), δ" t "l is Kurofutsuka's delta function, and δn * m H=1(" = mI
) v δn , m H= 0(n”ml).

The LPG composite film is driven by this driving sound source d (n) and outputs a composite signal x (m).

Assume that an all-pole digital filter is considered as an LPG synthesis filter, and its transfer function is expressed by the total impulse response k(n) (0≦n≦M-1), then the synthesized signal x (n) is gray It is expressed by equation (4) of “.

4 In equation (4), d(z) represents the driving sound source 7. Next, the weighting with auditory correction for the error between the input audio signal x(n) and the composite signal ) is shown by the formula.

e, (n) = I X (n) −x (n))
4w (n) (5) Further, the square false ring can be derived from the equation (5) and expressed by the following equation (6).

(6) In equation (6), M indicates the number of samples in the section that minimizes the error, and is selected to have a length of one analysis frame, for example. The multipulse as the optimal sound source pulse range can be obtained by obtaining gl that minimizes equation (6), and this g is determined by the above (3), (
The following equation (7) is derived from equations (4) and (6).

M=1 E hW(n-lTl, ) -b, (n-ml)n=
1 ・・・・・・・・・・・・(7) In equation (7), xw(”) idx(n)Nw(n)
,h,.

(n) indicates h(n)%w(n). The first term in the numerator on the right side of equation (7) indicates the cross-correlation function φhx (town) of the time delay ml between x, (n) and bW(n), and the second term E hvr(”-mt) ・iw(n
"s) beam, (n) M=1 covariance function φhh(mtymt) (141nzymt
≦M). The covariance function φhh (mz*m+) is equal to the one-autocorrelation function hhh (l mz −m Il ), and the following equation (7) can be expressed as the following equation (8).

(8) According to equation (8), if a pulse is generated at time position mI, the amplitude gI(m+) will be determined to be optimal. Note that in equation (8), 1-m1≦M
It is.

In other words, by focusing on a certain sound source pulse and calculating its amplitude using equation (8) at the time position of frame extraction, the one that maximizes the absolute amplitude of the amplitude minimizes the squared error shown in equation (6). By repeating this procedure, a plurality of sound source pulses can be generated.

Regarding the above-mentioned calculation algorithm, please refer to Obama, Hama, and Ono's study of multi-pulse canter speech coding method.
It was detailed in the Communication Method Study Group of the Institute of Electronics and Communication Engineers, March 1983.

By generating multi-pulses based on such a calculation algorithm, it is possible to calculate an optimal multi-pulse from the cross-correlation function, autocorrelation numerical value, and maximum value calculation, so the configuration can be made into a non-A1 configuration. This makes it possible to realize a multi-pulse vocoder that can be integrated into a single element, greatly reducing the amount of calculation required.

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

In other words, when the word length of the synthesis filter on the synthesis side is lentic, the synthesis filter 8 force using the multi-pulse driving sound # determined by the above-mentioned calculation algorithm will be effective when the word length of the synthesis filter on the synthesis side is lentic. When the threshold is small, noise power due to the limit cycle generated due to the effect of the visible word length of the synthesis filter interferes with the noise power, which has the disadvantage that it becomes extremely aurally harsh.

(Object of the Invention) The object of the present invention is to eliminate the above-mentioned drawbacks, and in a multi-pulse type vocoder, add a DC signal to the input audio signal on the analysis side to adjust the amplitude and position of the multi-pulse. By supplying a multi-pulse driving sound source that can generate a synthesized power sufficiently larger than the noise power due to the limit cycle of the synthesis filter, it is possible to generate sound even when the power of the voice is low, such as when there is no head or at the end of a voiced speech. An object of the present invention is to provide a multi-pulse vocoder that can generate good synthesized sounds.

(Structure of the Invention) The multi-pulse vocoder of the present invention performs LPG analysis on an input audio signal for each analysis frame, uses the extracted LPC coefficients as spectral envelope information, and configures the audio information of the input audio signal together with this spectral envelope information. The input audio signal is analyzed and synthesized by representing the sound source information in each analysis frame using a plurality of impulse system pulses (multipulses) having generation time positions and pulse widths corresponding to the characteristics of the sound source information. In multi-pulse vocoders.

means for adding a DC signal to the input audio signal; and means for calculating a cross-correlation coefficient sequence between the input audio signal to which the meandering signal has been added and the impulse response of the speech synthesis filter;
means for calculating an autocorrelation coefficient sequence of the impulse response; and means for forwardly calculating the amplitude and position of the impulse system 9I (multipulse) based on the relationship between the cross-correlation coefficient sequence and the autocorrelation. is configured on the analysis side.

(Example) Next, the present invention will be described in detail with reference to the drawings.

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 LPG analyzer 7. Cross-correlation function calculator8. Encoder (1)9. Autocorrelation function calculator 10, multipulse calculator 11. R encoder (2) 12° DC signal generator 1
It includes a 9% DC adder 2o and a multiplexer 13.

The input audio signal input via the input terminal 7001 is
It is supplied to a PC analyzer 7 and a DC adder 15.

The LPC analyzer 7 analyzes the input audio signal for each analysis frame.
Quantize as a digital acid with a preset number of bits,
This child-respected speech signal is subjected to LPG analysis to extract a p-order parameter (partial autocorrelation coefficient) as an LPC coefficient, and this rL is supplied to the encoder (1) 9 via an output line 701. In this example, the analysis frame is 20 m S
It is set to EC.

The encoder (1) 9 quantizes and encodes the input LPG coefficients, and then sends them to the multiplexer 13 via an output line 901.

LPG analyzer 7 is also. The impulse response h(n) (04n≦M-1) is calculated from the LPC coefficient, and the output line 70
2. Encoder (1)9. It is supplied via an output line 902 to a cross-correlation function calculator 8 and an autocorrelation function calculator 10.

The DC signal generator 19 is designed to generate a DC signal 9,
The amplitude of the generated DC signal is predetermined empirically, corresponding to the maximum amplitude of the voiced footstep of the input audio signal. In this embodiment, the DC signal generator 19 generates a curved signal having an amplitude 30 dB lower than the maximum amplitude of the input audio signal supplied via the input terminal 7001 and outputs it to the DC adder 20. The meandering adder 20 adds a DC signal to the input audio signal supplied via the input terminal 7001 and outputs it to the cross-correlation function calculator 8 .

The cross-correlation function calculator 8 calculates the cross-correlation function φ using the input audio signal to which the DC signal has been added and the impulse response: h(n), and calculates the cross-correlation function It is sent to the pulse calculator 11.

Further, the autocorrelation function calculating unit 10 calculates the autocorrelation function R5hh of the input impulse response h (n), and
ttl-similarity calculator 11 via output line 1001
Send to.

The multi-pulse calculator 11 calculates the cross-correlation function φ and autocorrelation function RhhI! for each analysis frame thus input. :
A predetermined number of sound source pulses 91I are obtained using the method described below using the After encoding and encoding, the multi-pulse? Send on 13th.

In this way, the LPC coefficients and multipulse data that are quantized and encoded and sent to the multiplexer 13 are processed in a predetermined manner via the multiplexer 13 as data representing the spectral envelope and sound source information of the input audio signal. The signal is time-divided and transmitted from the analysis side shown in FIG. 2 to the synthesis side shown in FIG. 3 via a transmission path 1301.

The synthesis side shown in FIG. 3 synthesizes input audio signals based on data transmitted from the analysis side via the transmission path 1301i, and has a demultiplex ratio of 14. Decoder (
1) 15. Discretion Ihi device (2) 16° LPC synthesizer 17
and L P F (Low Pa5sFilter
) 18 mag.

The multiplexer 14 restores various data input via the transmission line 1301 to the state before conversion by the multiplexer 13 in the time division transmission format, and the LPC rest field data is sent to the decoder (1) via the output line 141. ) 15, the multi-pulse theta is sent to the decoder (2) via output line 142.
) 16, the data is decoded by these decoders, and then sent to output lines 151 and 161, respectively.

The LPG synthesizer 17 uses the thus inputted multipulses as sound source information as a driving sound source for a p-order all-pole digital filter, and also uses the p-order LPC coefficient data input via an output line 151. is used as the coefficient of the above-mentioned all-pole digital filter to control this LPC synthesis filter to synthesize the input audio signal, which is sent to [, PF'18 via the male line 211, and is subjected to predetermined low-pass filtering for all rows of analog signals. A:' is sent to the Kaline 181 as a synthesized voice of the amount.

The sound source information expressed by the multi-pulse input to the LPC synthesizer 17 described above is at least 39 dB of the voiced sound constant part.
It has relatively low power, and must be sufficiently larger than the power generated by the limit cycle of the LPC synthesizer 17. Note that the DC component synthesized by the L1'C synthesizer 17 is not perceived audibly at all.

Next, a method applied to the multi-pulse calculator 11 will be explained. The multi-pulse Lin voice generator 11 calculates the above-mentioned cross-correlation coefficient φ and auto-correlation coefficient Rhh, l! for each analysis frame. :
Any method can be used as long as it can calculate a predetermined number of dorsal pulse trains using all the methods. - As an example, a case will be described in which the multi-pulse calculation method of the multi-pulse calculator 11 uses the above-mentioned small Paris algorithm.

First, the maximum absolute value of the cross-correlation coefficient φ is searched. Next, determine the lth f-pulse having the position and amplitude (change the polarity) of φhx found in step 11.

In order to remove the components of the long pulse, the
The autocorrelation of t "0" f is normalized by the number of threads rL is attached to 0 according to the amplitude (what polarity) of the first sound source pulse fc Rhh VC processing is performed to make the searched position Rhh
(0) and correct φbx. Next, the maximum absolute value of the corrected φbx is searched, a second sound source pulse is determined, and φhx is corrected. Civil engineering operation fees will be repeated as necessary.

Note that as a multipulse calculation method that does not rely on Obama et al.'s algorithm, there is a method that relies on similarity.

The method based on similarity searches for the maximum similarity between φbx and FLhh, such as the cross-correlation coefficient, instead of searching for the maximum absolute value of φhx, and is disclosed in patent #
This is the method described in No. 458-149007.

Further, in the above explanation, it was assumed that the perceptual weighting shown in equation (1) was not performed, but it is also possible to perform the perceptual weighting on the qBp/sensation. When perceptual weighting is carried out, a filter having a transfer function shown in equation (1) is configured with γ=0.8, for example, and inserted between the free-flow adder 20 and the cross-correlation function calculator 8. Instead of the impulse response calculated by the LPC analyzer 7 described in 1.
L with the damping coefficient r (γ=0.8) applied to the LPG coefficient
It is clear that the impulse response h(n)' calculated from the PC divisor can be used.

In the embodiments of the present invention shown in FIGS. 2 and 3, parameters are used as LPC coefficients, but other LPCI thread counts, such as the α parameter, etc. may also be used. It is clear that the encoder and multiplexer, and the decoder and the multiplexer, respectively, can be similarly implemented with these t-n configurations, and the LPG synthesis filter can also be implemented with a non-polar type other than an all-pole type. It is also clear that the implementation can be carried out in almost the same way even if a digital filter or the like is substituted.

(Effects of the Invention) As explained above, according to the present invention, in a multi-pulse vocoder, there is provided a means for adding a first-class signal to a human-powered speech signal, and a means for adding a first-class signal to a human-powered speech signal, and a means for adding a first-class signal to a human-powered speech signal, and an input speech signal to which the signal has been added and a speech synthesis filter. Based on means for calculating a cross-correlation coefficient with an impulse response, means for calculating an autocorrelation coefficient of the impulse response, and the relationship between the cross-correlation coefficient and the autocorrelation coefficient d1 of 111. Impulse system 911 (
By having a means for forward-calculating the image width and position of (multi-pulse) on the analysis side, the input sound source level of the LPC synthesizer can be summed up to 13y, It has the effect of making it sufficiently larger than the limit cycle of the instrument, making it possible to produce a good synthesized sound.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing the basic configuration 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 configuration of a conventional multi-pulse vocoder. FIG. 2 is a block diagram showing an embodiment of the synthesis side of a pulse-type vocoder. 1...LPC synthesizer, 2... Group LPC analyzer, 3... Sound source pulse generator, 4... Subtractor, 5... Old... Auditory weighting device , 6...2 error minimizer, 7...LPC analyzer, 8...
- Cross-correlation function calculator, 9... Encoder (1)
, lO... Autocorrelation function calculator. 11...Multi-pulse stretcher, 12...
・Encoded cotton (2), 13...Multiplexer,
14...Demultiplexer, 15...
Decoder (1), 16...Decoder (2), 1
7...LPC synthesizer, 18...LPF
, 19... Curved flow signal generator, 20...
Curved adder. thing 2 kunifu 3 old

Claims (1)

[Claims] The input audio signal is subjected to LPC (Linea) for each analysis frame.
r Prediction Coefficient,
The LPC coefficients analyzed and extracted (linear prediction coefficients) are used as spectral envelope information, and together with this spectral envelope information, the sound source information that constitutes the audio information of the input/output voice code is generated corresponding to all the features of this sound source information for each analysis frame. In a multi-pulse vocoder that analyzes and synthesizes the input audio signal by expressing it with a plurality of impulse sequences (multipulses) having time positions and amplitudes, means for adding a DC signal to the input audio signal. A multi-pulse vocoder characterized in that it has a function of determining the amplitude and position of a multi-pulse on an analysis side.