JPH043876B2 - - Google Patents

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.

(Prior art) An input audio signal is analyzed, spectral envelope information and sound source information that constitute the audio information of the input audio signal are extracted on the analysis side, and these audio information are sent 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 signal, and usually includes the number of LPC coefficients corresponding to the analysis order obtained by LPC analysis, such as the α parameter and the K parameter. The sound source information indicates the fine structure of the spectral envelope, and is known as the so-called residual signal obtained by removing the spectral distribution information from the input audio signal. It is also well known that information regarding pitch period and voiced/unvoiced is included, and that this information is usually extracted via autocorrelation for each analysis frame of the input audio signal.

When synthesizing input audio signals on the synthesis side of a vocoder, spectral envelope information is usually used as coefficients for an LPC synthesizer that uses all-polar digital film to form an approximate 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.

Although the conventional LPC vocoder obtained in this way is capable of synthesizing speech even at low bit rates of about 4Kb (kilowatts) or less and is widely used, it has the disadvantage that high-quality speech synthesis is difficult even at high bit rates. has. The reason for this is that when modeling sound source information, for a voiced sound, the pitch period corresponding to its content is extracted and approximately represented by a single impulse train corresponding to this pitch period, while unvoiced sounds with a random period are is assumed to be a simple modeling process in which it is approximated by white noise, so it does not faithfully extract the sound source information of the input audio signal, and therefore the input audio signal contained in the sound source information This is due to the fact that analysis and synthesis of waveform information has not been carried out.

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

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

The LPC synthesizer 1 is equipped with an all-pole digital filter that simulates the vocal tract, and its coefficients are input to the input terminal.
The input audio signal x(n)(n
= 1, 2, 3...n) by the LPC analyzer 2 for each analysis frame. The sound source pulse generator 3 obtains a driving sound source sequence V(n) consisting of a plurality of impulse sequences, that is, multipulses, from the sound source information of the input audio signal,
This is supplied as a driving sound source to the LPC synthesizer 1.

The LPC synthesizer 1 inputs the LPC coefficients in this way,
Usually, this is the coefficient of a synthesis filter using an all-pole digital filter, which is driven by a multi-pulse as a driving sound source and outputs a synthesis signal x〓(n). In this case, the multi-pulse includes waveform information of the input audio signal, and the LPC synthesizer 1 synthesizes the input audio signal including the waveform information.

Now, the composite signal x output from LPC synthesizer 1
(n) is then subtracted from the input audio signal x(n) by a subtracter 4 to obtain an error e(n), which is sent to the auditory weighter 5.

The perceptual weighting device 5 applies perceptual weighting to the error e(n) using a weighting filter having a characteristic w(z) shown in the following equation (1), and
These are sent to the square error minimizer 6.

w (z) = [1- _p 〓 ^k=1 a _k z ^-k ] / [1- _p 〓 ^k=1 a _k r ^k z ^-k ] ...(1) In equation (1), a _k is LPC The LPC coefficient to be used as the coefficient of the all-pole digital filter of synthesizer 1, p is its order and therefore the LPC analysis order, r is the weighting coefficient, and z is the transfer function H( z ⁻¹ ), where λ=2πΔT _f , ΔT is the sampling period of the analysis frame, and is the frequency.

In addition, in equation (1), the weighting coefficient r is 0<r<
It is set in the range of 1.

w(z) shown in equation (1) is 1 for r=1, r
= 0, the value of r varies within the range of w(z) = 1-p(z), and the value of r is set to an extent that suppresses the excessive level appearing in the formant region in the frequency spectrum of the error e(n). It is correspondingly set within the above-mentioned range and plays the role of audible weighting of the signals to be combined, and its optimum value is usually selected in advance by an optimum audibility test.

The error e(n) weighted in this way is
In order to determine the driving sound source sequence V(n) output from the sound source pulse generator 3, that is, the optimal time position and amplitude of the multi-pulse, it is sent to the square error minimizer 6, and is calculated according to the following equation (2). Calculate the squared error ε, ε
The driving sound source sequence V(n) is selected so as to minimize the value of V(n).

ε= _N 〓 ^n=k [e(n)*w(n)] ² ...(2) In equation (2), the symbol * is the convolution integral by the weighting filter of the auditory weighter 5, and N is the multipulse calculation Indicates the length of the interval.

The above-mentioned process is repeated for each multi-pulse, and synthesis by analysis is performed for each multi-pulse, which is the so-called Analysis-by-Syntonesis method (hereinafter abbreviated as A-b-S method).
As is clear from the above, this A-b-S method is an effective method in the low bit rate region because it is performed in a loop of pulse generation, square error calculation, and pulse position/amplitude adjustment for each multipulse. However, the disadvantage is that the amount of calculation required is extremely large.

Regarding this A-b-S method, BS
Atal el al, “A New Model of LPC
Excitation for Producing Natural−Sounding
Speech at Low Bit Rates”，Proc.ICASSP
82, pp. 614-617, (1982), etc.

In order to address these shortcomings in the conventional A-b-S method, the following arithmetic processing algorithm has recently been introduced which efficiently calculates optimal multi-pulses based on correlation calculations.

That is, the input audio signal x(n) is divided into processing frames every N samples, and multipulses are comprehensively calculated for each frame.

Assume that there are k sound source pulses in one analysis frame, and that the i-th pulse is at a time position m _i from the frame end and its amplitude is g _i , then the driving sound source d of the LPC synthesis filter is (n) is expressed by the following equation (3).

d(n)= _k 〓 ⁱ⁼¹ g _i・δn,m _i ...(3) In equation (3), δn,m _i is the Kronetzker delta function, and δ,m _i =1(n=m _i ), δn, m _i =0
(n≒ _mi ).

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

As an LPC synthesis filter, let us consider, for example, an all-pole digital filter, and its transfer function is expressed by an impulse response k(n) (0nM-1), then the synthesis signal x〓(n) is as follows (4 )
It is expressed by the formula.

x〓(n)= _M-1〓l ⁼⁰ d(l)・h(n-l)...(4) In equation (4), d(l) represents the driving sound source. Next, the weighting error obtained by performing auditory correction on the error between the input audio signal x(n) and the composite signal x〓(n) is e _w
(n), e _w (n) is expressed by the following equation (5).

e _w (n)={x(n)-x〓(n)}＊w(n)...(5) Furthermore, the squared error can be derived from equation (5) and expressed by the following equation (6). I can do it.

_M 〓 ⁿ⁼¹ e ² _w (n)= _M 〓 ⁿ⁼¹ [{x(n) −x〓(n)}*w(n)] ² ...(6) In equation (6), M is the error The number of samples in the interval that minimizes is selected, for example, as the length of one analysis frame. Multipulse as the optimal sound source pulse train is
Obtained by obtaining g _i that minimizes equation (6),
This g _i is derived from the above-mentioned equations (3), (4), and (6) as shown in equation (7).

g _i (m _i )= _M 〓 ⁿ⁼¹ x _w (n)・h _w (n−m _i ) − _i−1 〓 ^l=1 [gl _M 〓 ⁿ⁼¹ h _w (n−m _l )・h _w (n-m _i )] / _M 〓 ⁿ⁼¹ h _w (n-m _i )・h _w (n-m _i ) ...(7) In equation (7), x _w (n) is converted to x ( n) * w (n), h _w
(n) indicates h(n)*w(n). The first term in the numerator on the right side of equation (7) is the time delay between x _w (n) and h _w (n).
It shows the cross-correlation function φ _hx (m _i ) of m _i ,
Also, the second term _p 〓 ^k=1 h _w (n-m _l )・h _w (n-m _i ) is
_Covariance function φ _hh (m _l , m _i ) (1m _l , m _i
M) is shown. The covariance function φ _hh (m _l , m _i ) is equal to the autorelation function R _hh (|m _l −m _i |), so
Equation (7) can be expressed as the following equation (8).

According to equation (8), if a pulse is generated at time position m _i , the amplitude g _i (m _i ) can be determined to be optimal. Note that in equation (8), 1
m _i M.

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

Regarding the calculation algorithm mentioned above,
Ozawa, Araseki, Ono, ``Study of multi-pulse driven speech coding method'', March 1983, detailed in the Communications Method Study Group of the Institute of Electronics and Communication Engineers.

Generating multipulses based on such calculation algorithms makes it possible to calculate optimal multipulses from cross-correlation functions, autocorrelation functions, and maximum value calculations, which greatly simplifies the configuration. Therefore, it is possible to realize a multi-pulse vocoder that can significantly reduce the amount of calculation.

However, even the multi-pulse vocoder improved in this way still has the following drawbacks.

In other words, when the word length of the synthesis filter on the synthesis side is finite, the output of the synthesis filter using the multi-pulse driven sound source determined by the calculation algorithm described above will be reduced when the power of the voice is low, such as during silence or at the end of a voiced sound. In most cases, noise power due to limit cycles generated due to the finite word length of the synthesis filter interferes with the signal, and the disadvantage is 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 obtain the amplitude and position of the multi-pulse, and then send the signal to the synthesis side. By supplying a multi-pulse driven sound source that can generate a synthesized power that is sufficiently larger than the noise power due to the limit cycle of the synthesis filter, it is effective even when the power of the voice is low, such as during silence or voiced endings. The object of the present invention is to provide a multi-pulse vocoder that can generate synthetic sounds.

(Structure of the Invention) The multi-pulse vocoder of the present invention performs LPC 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. A multi-pulse type that analyzes and synthesizes the input audio signal by expressing sound source information in each analysis frame as a plurality of impulse sequences (multipulses) having generation time positions and amplitudes corresponding to the characteristics of the sound source information. In the vocoder, means for adding a DC signal to the input audio signal, means for calculating a cross-correlation coefficient sequence between the input audio signal to which the DC signal is added and the impulse response of the speech synthesis filter; The analysis side includes means for calculating a correlation coefficient sequence and means for calculating the amplitude and position of an impulse sequence (multipulse) in a forward manner based on the relationship between the cross-correlation coefficient sequence and the self-correlation. be done.

In the present invention, by applying a DC component, limit cycles can be prevented, and even when the power of the audio signal is small, the generation of noise due to this can be prevented. Furthermore, since the applied signal is a direct current component, it is not perceived audibly on the receiving side. Note that if the power of the audio signal is low, it may be possible to amplify it, but in this case, for faithful transmission, information indicating that it has been amplified is sent to the receiving side, and the receiving side uses this information to It is necessary to attenuate the combined signal. However, this increases the amount of information to be transmitted, which contradicts the original purpose of transmitting audio signals by reducing the amount of information as much as possible.

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

The analysis side of the multi-pulse vocoder according to the present invention shown in FIG.
0, multipulse calculator 11, encoder (2) 12,
It is configured to include a DC signal generator 19, a DC adder 20, and a multiplexer 13.

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

The LPC analyzer 7 quantizes the input audio signal as a digital quantity with a preset number of bits for each analysis frame, performs LPC analysis on this quantized audio signal, and calculates the p-order K parameter (partial self-correlation) as an LPC coefficient. relation coefficient) and output this to the output line 701
The signal is supplied to the encoder (1) 9 via the encoder (1) 9. In this example, the analysis frame is set to 20 mSEC.

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

The LPC analyzer 7 also calculates an impulse response h(n) (0 nM-1) from the LPC coefficients, and outputs the result to the cross-correlation function calculator 8 and It is supplied to an autocorrelation function calculator 10.

The DC signal generator 19 generates a DC signal, and the amplitude of the generated DC signal is empirically determined in advance in accordance with the maximum amplitude of the voiced sound stationary portion of the input audio signal. In this embodiment, the input terminal
The DC signal generator 19 generates a DC signal with an amplitude 30 dB lower than the maximum amplitude of the input audio signal supplied via the input audio signal 7001 and outputs it to the DC adder 20 . The DC adder 20 adds a DC signal to the input audio signal supplied via the input terminal 7001 and outputs the result to the cross-correlation function calculator 8.

The cross-correlation function calculator 8 calculates a cross-correlation function φ _hx using the input audio signal to which the DC signal has been added and the impulse response h(n), and sends it to the multipulse calculator via an output line 801. Send it to 11.

Further, the autocorrelation function calculator 10 calculates the autocorrelation function _Rhh of the input impulse response h(n), and sends it to the similarity calculator 11 via the output line 1001.

The multi-pulse calculator 11 uses the cross-correlation function φ _hx and auto-correlation function R _hh for each analysis frame input in this way to obtain a predetermined number of sound source pulse trains using a method described later, and calculates the amplitude of these pulses. and position information are sent to encoder (2) 12 via output line 1101, where they are quantized and encoded, and then sent to multiplexer 13 via output line 1201.

In this way, the LPC coefficients and multipulse data that are quantized and encoded and sent to the multiplexer 13 are time-divided 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 transmitted from the analysis side shown in FIG. 2 to the synthesis side shown in FIG. 3 via the transmission path 1301.

The synthesis side shown in FIG. 3 synthesizes input audio signals based on data transmitted from the analysis side via a transmission path 1301, and multiplexer 1
4. It is configured with a decoder (1) 15, a decoder (2) 16, an LPC synthesizer 17, an LPF (Low Pass Filter) 18, and the like.

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

The LPC synthesizer 17 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 LPC coefficient data input via the output line 151 as the sound source information. This is used as the coefficient of an all-pole digital filter.
The LPC synthesis filter is controlled to synthesize input audio signals, which are sent to the LPF 18 via the output line 211, subjected to predetermined low-pass filtering, and sent to the output line 181 as analog synthesized audio.

The sound source information expressed by the multi-pulses input to the LPC synthesizer 17 described above has power at least about 30 dB lower than that of the voiced sound stationary part,
This is sufficiently larger than the power generated by the limit cycle of the LPC synthesizer 17. Note that the DC component synthesized by the LPC synthesizer 17 is not perceptible audibly.

Next, a method applied to the multi-pulse calculator 11 will be explained. The multi-pulse calculator 11 can use any method that can calculate a predetermined number of sound source pulse trains using the cross-correlation coefficient φ _hx and auto-correlation coefficient R _hh for each analysis frame described above. be. As an example, a case will be described in which the algorithm of Ozawa et al. is used as the multipulse calculation method of the multipulse calculator 11.

First, the maximum absolute value of the cross-correlation coefficient φ _hx is searched. Next, the first sound source pulse having the searched position and amplitude (having polarity) of φ _hx is determined. Furthermore, in order to remove the component of the determined sound source pulse, the search is performed by R _hh normalized by the autocorrelation coefficient of lag "0" and weighted by the amplitude (with polarity) of the first sound source pulse. φ _hx is corrected by associating the detected position with R _hh (0). Next, the maximum absolute value of the corrected φ _hx is searched, the second sound source pulse is determined, and φ _hx
Correct. Repeat the above operation as necessary.

Note that as a multipulse calculation method that does not rely on Ozawa et al.'s algorithm, there is a method that relies on similarity. The method based on similarity searches for the maximum similarity between φ _hx and R _hh , such as the cross-correlation coefficient, instead of searching for the maximum absolute value of φ _hx . This is the method described in 149007.

Further, in the above explanation, it is assumed that the perceptual weighting shown in equation (1) is not performed, but it is also possible to implement the perceptual weighting. When perceptual weighting is performed, a filter having a transfer function shown in equation (1) is configured, for example, with γ = 0.8, and is inserted between the DC adder 20 and the cross-correlation function calculator 8, and the filter is inserted between the DC adder 20 and the cross-correlation function calculator 8. It is clear that instead of the impulse response calculated by the analyzer 7, the impulse response h(n) calculated from the LPC coefficient to which the attenuation coefficient γ (γ=0.8) is applied may be used.

Note that in the embodiments of the present invention shown in FIGS. 2 and 3, the K parameter is used as the LPC coefficient, but this is different from other LPC coefficients, such as α
It is clear that the encoder and the multiplexer, and the decoder and the multiplexer can be similarly implemented as integrated configurations, and the LPC synthesis filter It is also clear that it can be implemented in almost the same way even if it is replaced with a non-polar type digital filter or the like other than the all-polar type.

(Effects of the Invention) As described above, according to the present invention, in a multipulse vocoder, there is provided a means for adding a DC signal to an input audio signal, and an input audio signal to which the DC signal has been added and an impulse response of a speech synthesis filter. means for calculating a cross-correlation coefficient of the impulse response, means for calculating an autocorrelation coefficient sequence of the impulse response, and calculating an impulse sequence (multipulse) based on the relationship between the cross-correlation coefficient sequence and the autocorrelation coefficient sequence. By having a means to calculate the amplitude and position in a forward manner on the analysis side, the input sound source level of the LPC synthesizer can be adjusted to a level higher than the limit cycle of the synthesizer even for voice parts with low power such as silence and voiced endings. This has the effect of making it possible to make the sound louder, and producing a better 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...LPC analyzer, 3...
... Sound source pulse generator, 4 ... Subtractor, 5 ... Auditory weighting device, 6 ... Square error minimizer, 7 ...
LPC analyzer, 8... Cross correlation function calculator, 9...
... Encoder (1), 10 ... Autocorrelation function calculator, 1
1...multipulse calculator, 12...encoder
(2), 13...Multiplexer, 14...Demultiplexer, 15...Decoder (1), 16...Decoder (2), 17...LPC combiner, 18...LPF, 1
9...DC signal generator, 20...DC adder.

Claims (1)

[Claims] 1 Analyze input audio signal by LPC for each frame
(Linear Prediction Coefficient)
The analyzed and extracted LPC 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 audio signal is analyzed, and the generation time position and amplitude corresponding to the features are calculated for each analysis frame. In a multi-pulse type vocoder that analyzes and synthesizes the input audio signal by expressing it with a plurality of impulse sequences (multi-pulses), the multi-pulse vocoder includes means for adding a DC signal to the input audio signal, A multi-pulse vocoder characterized by having a function of determining amplitude and position on an analysis side.