JPS627098A - Multipulse encoder - 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] [Field of Industrial Application] The present invention relates to a multipulse encoding device, and in particular has a function of performing gain adjustment on the output waveform of a speech synthesis filter, and synchronizes with the analysis frame period on the analysis side. A multi-pulse string code in a multi-pulse encoding device having a means for obtaining synthesized speech by linearly adding the speech waveforms independently synthesized for each analysis frame while maintaining continuity between the analysis frames and then linearly adding them. The present invention relates to a multi-pulse encoding device with improved coding efficiency.

[Conventional technology]

A multipulse encoding device is an encoding device that uses multipulses as sound source waveform information, and usually synthesizes audio waveforms using fixed-point finite-kind calculations. In addition, in speech synthesis, in order to improve the dynamic range and S/N (Signal/No1se) ratio of the speech synthesis filter as much as possible, all gains are performed on the output side of the filter in accordance with the strength of the sound source. In a multi-pulse encoding device such as
The signal is sent to the synthesis side together with spectral envelope information based on coefficients (e.g., @-type predictive coding) to perform speech synthesis.

These time-varying parameters sent from the analysis side to the synthesis side are transmitted in units of analysis frames or variable length frames, but in either case, speech synthesis is processed in units of analysis frames.

Now, the analysis frame mentioned above is the analysis of feature parameters,
The length of time is set from the viewpoint of transmission convenience, and therefore it is basically unrelated to faithful transmission of characteristic parameters.For this reason, it is assumed that the sound source waveform is transmitted using a multi-pulse train. Even in multi-pulse encoding devices, it is difficult to ensure faithful continuity of the synthesized waveform across analysis frame boundaries by simply continuously using such uniform feature parameters for each analysis frame on the synthesis side. There is a problem in that it is ignored and the waveform reproducibility inevitably deteriorates.

To address this problem, the following improvement measures have recently been introduced.0 In other words, speech synthesis of multi-pulse trains while synchronizing with the analysis frame on the analysis side and maintaining continuity between analysis frames. This is a multi-pulse encoding device that basically eliminates the problem of discontinuity between analysis frames by reproducing a composite waveform as a summation of impulse response waveforms by a filter while taking into account the impulse response duration. be. The present invention is also directed to such a multipulse encoding device.

[Problem that the invention seeks to solve]

However, even with this conventional multipulse encoding device by Hajiashi, which eliminates the problem of discontinuity between analysis frames, the following problems unique to the multipulse encoding device remain.

In other words, multi-pulse is a search for a linear summation of impulse responses of a speech synthesis filter configured as an all-pole model that approximates as much as possible the input speech shaping, and therefore a multiple pulse train that approximates the f# waveform. is being searched as. There are extreme y5 methods for this search, but in any case, it is configured as a multiplication of many pulses that can form a plurality of excitation points corresponding to the vocal cord imaging waveform 11.
Therefore, there is a problem that the coding efficiency is limited.The purpose of the present invention is to eliminate the above-mentioned drawbacks and to combine multiple excitation points into one.
A multi-pulse encoding device that is equipped with a means to search for a multi-pulse having an excitation point that can be represented by a fake or at most several 11 m, thereby greatly reducing the number of multi-pulses and greatly improving their encoding efficiency. Our goal is to provide the following.

[Means for solving 1 and 4 points] The device of the present invention has a function of performing gain adjustment on the output waveform of the speech synthesis filter, and is synchronized with the analysis claim period of the analysis side and is continuous between analysis frames. In a multipulse encoding device having a means for obtaining speech synthesis by linearly adding the speech waveforms independently synthesized for each 7V-me while maintaining their characteristics, after adjusting the gain, multipulse analysis is performed and the search is performed. a phase equalization filter coefficient determining means for determining a filter coefficient of a phase equalization filter of an input back door signal for each analysis frame based on a multi-pulse train; and a multi-pulse determining means for determining the output multi-pulse by performing multi-pulse analysis again for each analysis frame. 0th 1st
The figure is a block diagram showing the configuration of an embodiment of the multi-pulse encoding device according to the present invention. The multi-pulse encoding device shown in FIG. is the window treatment device 101. Switcher 102. Phase conversion filter 103. Switcher 104. Coefficient calculator 1o5.
Multipulse analyzer 106.1. PC analyzer 107° multipulse encoder 108. LPC coefficient encoder 109
and a multi-breather 110. Furthermore, the 1 synthesis side 2 includes a demultiplexer 201, an LPC coefficient decoder 202, and so on. Normalized multipulse decoder 203. - Maximum amplitude decoder 2o4° Consists of LPC synthesis filter 2051, multiplier 206, and waveform connector 207.

The input audio signal input through the input line 1001 is passed through the window processor 101 through an LPF (Low Pass Pass).
The sample is sampled by an A/D converter at a predetermined sampling frequency of 81d (z), and then quantized for each sample at a predetermined number of 12 bits. This quantized audio signal is cut out and stored in the -H internal memory by a preset window function, for example, a rectangular function, for each preset time, for example, 20m5EC, that is, 160 samples.

The quantized audio signal thus stored is also 20m5EC
The signals are read out one after another and supplied to the multi-pulse analyzer 106 via switchers 102 and 104, and this becomes the basic analysis frame length in multi-pulse analysis.

Multipulse analysis 5106 extracts a multipulse train for each basic analysis frame using a known multipulse extraction technique.

As a known multi-pulse train extraction technique, AbS (Ana Iys i) is a so-called spectral domain evaluation.
5-bysynthesis) processing, correlation function processing as so-called correlation region evaluation, etc., and in this embodiment, the latter method is used.

This correlation region evaluation searches for a desired multipulse sequence by using the cross-correlation between the input audio signal and the impulse response of the analysis-evaluation synthesis filter on the analysis side and the autocorrelation of the impulse response. The analysis and evaluation synthesis filter used for this purpose is built into the LPC analyzer 107.
A multi-pulse search is performed while inputting an impulse response from . In this embodiment, this multi-pulse search is performed for a long period of time for each analysis frame, except for the first basic analysis frame, using a so-called isolated time that is preset by taking into account the impulse response duration due to multi-pulses. It is as follows.

The LPC analyzer 107 converts the quantized audio signal into 30rnSBC
While cutting out using the Hamming function of
PG analysis is performed to extract LPC coefficients of a predetermined order.

These LPC coefficients are used as coefficients of the built-in synthesis filter to obtain the auditory weighted impulse response of the vocal tract filter. This impulse response is multipulse analysis m106
supplied to

The multipulse analyzer 106 first searches for the multipulse train of the foremost basic analysis frame through a cross-correlation calculation between the quantized audio signal and the impulse response for each basic analysis frame and an autocorrelation calculation of the impulse response.

Now, the multi-pulse train of the foremost basic analysis frame obtained in this way is supplied to the LPC analyzer 107, and an impulse response is obtained by the built-in synthesis filter. The impulse response waveform generated by this multi-pulse is supplied to the multi-pulse analyzer 106 with both a portion spanning the basic analysis frame and a portion connecting beyond the basic analysis frame. However, the impulse response connection time beyond the basic analysis frame is limited to a time range set in advance, taking into consideration the quality level of the synthesized speech.

The multipulse analyzer 106 adds the quantized audio signal of the following basic analysis frame and the doji-ized data of the impulse response waveform of the previous basic analysis frame spanning the time domain of this basic analysis frame, taking polarity into account. Quantized audio signals that are synthesized and maintain continuity between basic analysis frames are set one after another for successive basic analysis frames. In this way, the multipulse train is searched again while maintaining the continuity between the basic analysis frames.

Although the multi-pulse train searched in this way maintains continuity between basic analysis frames, it is basically no different from the conventional multi-pulse encoding device from the viewpoint of multi-pulse train encoding efficiency. Therefore, in this embodiment, the following phase equalization is applied to the output of the window processing 5101 to solve this problem.

The basic idea of this phase equalization is as follows. In other words, there is a multi-pulse train that has just been retrieved, and when this is applied to a speech synthesis filter that can be composed of an all-pole digital filter consisting of LPCqf, its ・
Assume that linear addition of the impulse response waveforms provides the desired source 1# information. Such a digital filter can be easily constructed.

In this case, instead of the searched multipulse, a representative impulse is set based on these multipulse trains,
It is assumed that this impulse response can also be obtained that almost satisfies the sound source information. Such impulse settings can also be easily set.

On the other hand, if there is a filter that converts the impulse response waveform obtained by a multi-pulse train into the response waveform of a representative impulse (RC), then if the input audio signal is passed through this filter and then searched for a multi-pulse, the representative impulse is replaced by the multi-pulse train. It is clear that this is required. The purpose of phase equalization of the input audio signal is to use such a filter to obtain such a representative impulse with a significantly reduced number of pulses instead of a multi-pulse. Such a filter is a phase equalization filter. Note that a plurality of representative impulses described above are usually obtained for each basic analysis frame, corresponding to pitch periods and the like.

The basic concept of phase equalization filters was introduced by Mr. Honda 9, Sutani, for example, "Speech Coding Using Phase Equalization Processing," Acoustical Society of Japan Speech Study Group Material 584-05.1.
Written in April 1984.

Now, it is clear that the above-mentioned phase equalization filter can be easily realized by a transversal type digital filter or the like, and the one-phase equalization filter 103 is a digital filter for such purpose.

The coefficient calculator 105 provides filter coefficients for the phase equalization filter 103, and the filter coefficients are determined as follows using the multipulse train supplied from the multipulse analyzer 106.

Now, the multipulse (or predicted residual waveform) is expressed as e (n
), and the impulse response of the phase equalization filter is bi
n) (m=0, 1, 2.-, M), the phase equalization residual difference waveform can be expressed as ep(n) of the following equation (1).

Further, the pi,te pulse train is represented by eM(n) in equation (2).

δ in equation (2) is the delta of the Croneo force, and when n = nz, that is, δ(0), it becomes a pulse with an amplitude of 1, and nz'-
nt-t is the pitch period.

The coefficient calculator 105 inputs the multipulse e (n) and e
First, calculate the pitch period by finding the autocorrelation coefficient of (n) and searching for the maximum value of this coefficient. , the Teval pulse train eM(n) is selected from the multi-pulse train.

The desired h(e) which is best expressed by the above e p(n) and eM(n) is the phase equalized residual waveform e p(nl) shown by the equation (1) and the desired h(e) which is expressed by the equation (2). It is obtained by minimizing the squared error with the pulse train e M(n).

The squared error is expressed by the following equation (3).

...(3) However, in equation 3), L is the number of pitch pulses included in the interval 0 to M-IK.

By setting
・e(n-k)) ・・・・・・(4)(4
) is set to zero, the following equation (5) is obtained.

......(5) (k = 0.1, 2..., M) From equation (5), h- is obtained from the (k+x) element (k+1) order equation. This calculation formula is as follows. It can be expressed as a matrix equation as shown in equation (6).

...(6) In equation (6), Vβ) and φi8) are each (7)i
8) The content is shown in formula.

In equation (8), y(n) is a waveform representation of the pitch pulse train.

As is clear from equation (6), the coefficient matrix, x, is the autocorrelation coefficient of the multipulse, and the coefficient vector is the crosscorrelation coefficient of the multipulse and the pitch pulse train.

The coefficient calculator 105 calculates the coefficient coefficient of the transversal filter having this impulse response after performing the calculation based on the force (+6) and (8)2 and calculating the optimum impulse response coefficient. It is supplied as a filter coefficient of the phase equalization filter 103.

The phase equalization filter 103 is configured as a desired transversal type digital filter using the input filter coefficients. Phase equalization filter 10 configured in this way
The impulse response of 3 is of course htn).

The doji-ized audio signal for each basic analysis frame output from the window processor 101 is supplied to the phase equalization filter 103 via the switch 102, and its impulse response output is sent to the switch 1.
04 to the multi-pulse analyzer 106 and the LPC analyzer 107. The multi-pulse analyzer 106 thus receives the phase-equalized doji-ized audio signal for each basic analysis 7 layer from the phase equalization filter 103. A multi-pulse search is performed using correlation region evaluation in the same way as the processing described above.In this case, the LPC divider 107 also receives the same input from the phase equalization filter 103 for each basic analysis frame and converts it into an LPC The impulse response obtained by extracting the LPC coefficients of the order required for analysis and using them as coefficients of the built-in speech synthesis filter is provided to the multipulse analyzer 106. However, the quantized speech signal input to the LPC9pfr device 107 in this way Although it is input through the phase equalization filter 103, it is the same as above that basically maintains the sound source characteristics. However, it may be arbitrarily determined whether or not to perform the LPC analysis again.

In this way, the multipulse analyzer 105 determines the number of representative impulses set corresponding to the vi, te periods, etc. for each basic analysis frame, instead of the conventional multipulse, but the multipulse ( The number of representative multipulses (hereinafter referred to simply as representative multipulses) can be significantly reduced compared to conventional multipulses, and they also have the following characteristics.

In other words, in determining a more representative multi-pulse, the number of pulses in the multi-pulse train is reduced only by changing the frequency vs. phase characteristics without making any total thermal change to the electric flux vector of the input audio signal. The multipulse train consisting of representative multipulses analyzed and determined by the multipulse analyzer 106 is supplied to the multipulse encoder 108. Encoded in a predetermined format. The input of multi-pulses to the encoder 108 is performed at the cycle of the basic analysis frame, and this basic analysis frame is reset with the content of the impulse response duration taken into consideration as described above. The multipulse train for each reset basic analysis frame has a maximum amplitude code that encodes the amplitude of the maximum level one among them and its time position information, and a multipulse obtained for each reset basic analysis frame as the maximum amplitude code. In this embodiment, the normalized multi-pulse code is divided into an amplitude value normalized using the amplitude as a reference and a normalized multi-pulse code related to time position information, and is supplied to the multiplexer 110. The LPC coefficient encoder 109 receives the LPG coefficients for each reset basic analysis frame from the LPC analyzer 107 and encodes them in a predetermined format to generate an LPC code. and supplied to Shite Marte Grexa 110.

The multiplexer 110 multiplies each encoded data received in this way in the form of pruning and transmits it to the combining side 2 via the transmission line 1002. On the combining side 2, the demultiplexer 201 separates the input multiplicity, and then the LPG code is converted into LPG code. The normalized multipulse code is sent to the coefficient decoder 202, the normalized multipulse code is sent to the normalized multipulse decoder 203, and the maximum amplitude code is sent to the maximum amplitude decoder 20.
4 and are decoded.

The LPC coefficients thus decoded are supplied to the LPC synthesis filter 205 and used as filter coefficients.

Further, the normalized multi-pulse is supplied as a driving sound source to the LPC synthesis filter 205, and the maximum amplitude is supplied to the multiplier 205.
6, the LPC synthesis filter 205 is supplied to the LPC analyzer 10 on the analysis side.
It is basically an all-pole type digital filter that is almost the same as the built-in analytical synthesis filter of No. 7, and uses LPG coefficients as filter coefficients and is driven with normalized multi-pulses as sound source information, and outputs a digital audio signal. However, multipulse amplitude information is not yet used in this digital audio signal.

LPG synthesis filter 205 in AN LPC vocoder etc.
The maximum amplitude decoded on the 0 input side is used to restore normalization and create a multi-pulse train with actual level information added. In this embodiment, this is performed at the output 1111 of the LPC synthesis filter 205.

The digital audio signal output from the LPC synthesis filter 205 is supplied to the multiplier 206 for each reset basic analysis frame.

Maximum amplitude data is also supplied to the multiplier 206 for each reset basic analysis frame, and the digital audio signal to which actual level amplitude information has been added through these two manual multiplications is supplied to the waveform connector 207.

In this way, the waveform connector 207 linearly adds the digital audio signals that are input one after another for each resetting analysis frame in time series, converts them into analog audio signals using the 1)/A converter, and roughly passes them through the LPF to eliminate unnecessary signals. high frequency components are removed and sent as output audio to the output line 2001.
In this way, speech synthesis through the process of independently synthesizing speech waveforms for each analysis frame while synchronizing with the analysis frame period on the analysis side and maintaining continuity between analysis frames greatly improves multipulse train coding efficiency. This will be implemented at each station under the

FIG. 2 is a schematic diagram showing the configuration of a second embodiment of the present invention, which consists of an analysis side 1' and a synthesis side 2', and the analysis side 1' has a window processor 101. Switcher 102, phase equalization filter 103. Switcher 104° Coefficient Calculator 105. Multipulse analyzer 106° LPC analyzer 107. Multipulse encoder 108° LPC coefficients 109. It is configured to include a multiplexer 110 and an equalization coefficient encoder 111, and the synthesis side 2' includes a demultiplexer 201. LPC coefficient encoder 202. Normalized multipulse decoder 203
, maximum amplitude decoder 204 . LPC synthesis filter 205
1 multiplier 206. It is configured with a waveform connector 207, an equalization coefficient decoder 208, and a phase dispersion filter 209.

Among these components, the analysis side i'i'c is the equalization coefficient encoder 111, and the synthesis @ll2' is the equalization coefficient post-encoder 208 and the phase dispersion filter 209. They are exactly the same as those with the same symbols on the analysis side 1' and the synthesis side 2' in the first embodiment shown in the figure.
Detailed explanations regarding these same items will be omitted. This second embodiment encodes the phase equalization filter coefficients calculated on the analysis side and transmits them to the synthesis side together with the Martenorculus, LPC coefficients, etc. , which attempts to further improve the quality of synthesized speech by performing phase dispersion of speech waveforms synthesized independently for each analysis frame.

The equalization coefficient encoder 111 on the analysis side 1' receives data having the same content as the phase equalization filter coefficients supplied to the phase equalization filter 103 from the coefficient calculator 105, encodes it in a predetermined format, and then encodes it in a predetermined format. Q to be sent to the multiplexer 110 The multiplexer 110 multiplexes the equalization coefficient code together with the maximum amplitude code, the normalized multipulse code, and the I and PC codes in a predetermined format and sends it to the transmission path 100.
2 to the combining side 2'. On the combining side 2', the demultiplexer 201 demultiplexes the received signal and the equalization coefficient code is sent to the equalization coefficient decoder 2.
The equalization coefficients decoded by the equalization coefficients are supplied to the phase dispersion filter 209.

The phase dispersion filter 209 is a digital filter configured as a transversal filter, and uses equalization coefficients as its filter coefficients to perform inverse processing of the equalization processing performed on the input audio signal by the phase equalization filter 103 performed on the analysis side. Phase dispersion processing using LPC synthesis filter 2
After resetting the output of 05 for each basic analysis frame, the output is supplied to the multiplier 206, and synthesized speech is obtained in the same manner as in the first embodiment.

The purpose of achieving full phase dispersion on the synthesis side in this way is to eliminate the effects of phase equalization performed on the analysis side, with the aim of greatly improving the encoding efficiency of multipulse sequences, and This can be carried out arbitrarily, taking into consideration the quality conditions of the synthesized speech for the purpose of operation of the device.

〔Effect of the invention〕

As explained above, according to the present invention, it has the function of performing gain adjustment on the output waveform of the speech synthesis filter, and
A multipulse encoding device that obtains synthesized speech by linearly adding the speech waveforms synthesized independently for each analysis frame while synchronizing with the analysis frame period on the analysis side and maintaining continuity between the analysis frames, after adjusting the respective gains. In this method, the number of simultaneous pulses can be greatly reduced and the multipulse analysis can be performed again after determining the coefficients of the phase equalization filter from the retrieved multipulse train and phase equalizing the input audio for each analysis frame. This has the effect of realizing a multi-pulse encoding device with significantly improved pulse encoding efficiency.

[Brief explanation of drawings]

The IF diagram is a diagram showing the configuration of the first embodiment of the present invention, and FIG. 2 is a diagram showing the configuration of the second embodiment of the invention. 1.1'... Analysis side, 2.2'... Synthesis side, 101... Window processor, 102...
Switcher, 103...Phase equalization filter% 10
4...Switcher, 105...Coefficient calculator, 1
06...Multipulse analyzer, 107...
・LPC portion 4ff5. l Os...Multipulse encoder, 109...L, PC coefficient encoder, 110...Multiplexer, 111...
... Equalization coefficient encoder, 201 ... Demultiplexer, 2o2 ... LPC coefficient ei decoder,
203... Normalized multipulse decoder, 204
...Maximum amplitude decoder. 205...LPC synthesis filter, 206...
... Multiplier, 207 ... Waveform connector, 208
. . . Equalization coefficient decoder, 209 . . . Phase dispersion filter. Kaya! Concave number 21! ! ]

Claims (2)

[Claims] (1) It has a function to perform gain adjustment on the output waveform of the speech synthesis filter, and is synthesized independently for each analysis frame while synchronizing with the analysis frame cycle on the analysis side and maintaining continuity between analysis frames. In a multi-pulse encoding device having means for obtaining synthesized speech by linearly adding each waveform after adjusting the gain thereof, a phase equalization filter is applied to the input speech signal for each analysis frame based on the multi-pulse string retrieved by performing multi-pulse analysis. a phase equalization filter coefficient determining means for determining a filter coefficient of the filter; A multi-pulse encoding device comprising: multi-pulse determining means for determining a multi-pulse. (2) The filter coefficients of the phase equalization filter are transmitted from the analysis side to the synthesis side, and the phase dispersion of the speech waveform synthesized independently for each analysis frame is performed so as to eliminate the effect of the phase equalization. The multi-pulse encoding device according to item (1).