JP2615862B2 - Voice encoding / decoding method and apparatus - Google Patents

Abstract

PURPOSE:To improve sound quality and to drastically decrease the quantity of computation by obtaining multipulses without using pitch prediction for the pulses which are not strongly correlated with each of pitches. CONSTITUTION:The sound source signal of the voice signal of a frame section is indicated by using the 1st multipulses obtd. by pitch prediction and the 2nd multipulses obtd. by without pitch prediction. Further, the frame is previously divided to the sub-frames corresponding to the pitch periods and the multipulses are obtd. by the pitch prediction only with the one section of the sub-frames in order to utilize the quasiperiodicity of every pitch of the multipulse sound source and to decrease the computation quantity. The signal is reproduced by the multipulses to remove the influence thereof; thereafter, the multipulses are obtd. without the p[itch prediction in said frame. Namely, a 1st pulse decoder 720 decodes the amplitude and position of the pitch predicted multipulses and a 2nd pulse decoder 725 decodes the amplitude and position of the 2nd multipulses.

Description

Description: TECHNICAL FIELD The present invention relates to an audio encoding / decoding method and apparatus for encoding and decoding an audio signal efficiently at a low bit rate.

(Prior Art) As a method of transmitting an audio signal at a low bit rate, for example, about 16 Kb / s or less, a multi-pulse encoding method and the like are known. In these, the sound source signal is represented by a combination of a plurality of pulses (multi-pulse), the characteristics of the vocal tract are represented by a digital filter, and the information of the sound source pulse and the coefficient of the filter are represented by:
It is obtained and transmitted for each fixed time section (frame). For details of this method, for example, Araseki, Ozawa, Ono, Oc
Hiai “Multi-pulse Excited Speech Coder Ba
sed on Maximum Cross-correlation Search Algorith
m ", (GLOBECOM83, IEEE Global Tele-communication, Lecture No. 23.3, 1983) (Reference 1). In this method, vocal tract information and a sound source signal are separated and expressed, respectively. By using a combination of a plurality of pulse trains (multi-pulse) as a means for expressing a sound signal, a good speech signal can be output after decoding. The basic concept of obtaining a pulse train representing a sound source signal will be described with reference to FIG. An audio signal divided for each frame is input from an input terminal 800 in the figure.
In 0, a spectrum parameter obtained from the audio signal of the current frame is input. An initial multi-pulse is generated in a sound source calculation circuit 810, and this is
By inputting to 820, a synthesized speech waveform is obtained as an output. The subtractor 840 subtracts the synthesized speech waveform from the input signal. The result is input to the weighting circuit 850 to obtain the weighted error power in the current frame. Then, the amplitude and position of a specified number of multi-pulses are obtained in the sound source generation circuit 810 so as to minimize the weighted error power.

(Problems to be Solved by the Invention) However, in this conventional method, the sound quality is good when the bit rate is sufficiently high and the number of sound source pulses is sufficient, but the sound quality deteriorates as the bit rate is reduced. There was a problem.

In order to improve this problem, a pitch prediction multipulse method using quasi-periodicity (pitch correlation) for each pitch of a multipulse sound source has been proposed. For more information on this method,
For example, since it is detailed in Japanese Patent Application No. 58-139022 (Reference 2), the description is omitted here. However, the quasi-periodicity of each pitch of a multi-pulse sound source is considered to be large for large-amplitude pulses, but not all pulses have such periodicity. Is considered to be small. In the pitch prediction multi-pulse method described in Reference 2, all pulses are obtained by pitch prediction assuming periodicity for each pitch for all predetermined numbers of pulses in a frame. In the case of the pulse, there is a problem that the characteristics are deteriorated by the pitch prediction. In particular, this is remarkable in the transition section and transition part of vowels,
There is a problem that sound quality is deteriorated in such a portion.

Further, in the method of Reference 2, since pitch information is included in the impulse response, an extremely long impulse response (for example, 20 to 32 ms) is required, and a predetermined number of all the pulses are subjected to pitch prediction. Therefore, the amount of calculation required for searching for a pulse is very large, and it is quite difficult to realize a compact apparatus using current device technology.

SUMMARY OF THE INVENTION An object of the present invention is to provide a speech encoding / decoding method and apparatus capable of reproducing sound better than before even if the bit rate is high or lowering the bit rate and realizing it with a small amount of computation. Is to do.

(Means for Solving the Problems) According to a speech encoding / decoding method of the present invention, a discrete speech signal is input on the transmission side, and a spectrum parameter representing a spectrum envelope and a pitch representing a pitch are provided for each frame from the speech signal. Parameters, and divides the audio signal of the frame into small sections corresponding to the pitch parameter. For one of the small sections as a sound source signal, predicts the pitch using the spectrum parameter and the pitch parameter. Calculating the first multi-pulse while removing the influence of the first multi-pulse from the audio signal while performing the second multi-pulse without pitch prediction using the spectrum parameter in the frame, The spectral parameter, the pitch parameter, the first multipulse, and the second Transmitting a multi-pulse, reconstructing a sound source signal using the first multi-pulse, the pitch parameter, and the second multi-pulse on the receiving side, and obtaining a synthesized speech signal using the spectrum parameter. And

A speech encoding apparatus according to the present invention includes a parameter calculation circuit that extracts and encodes a spectrum parameter representing a spectrum envelope and a pitch parameter representing a pitch for each frame from an input discrete speech signal, and a speech signal of the frame. A dividing circuit that divides the signal into small sections corresponding to the pitch parameter; and a first multi-pulse while performing pitch prediction using one of the small sections as a sound source signal using the spectrum parameter and the pitch parameter. A sound source calculation circuit for obtaining and encoding a second multi-pulse without pitch prediction using the spectral parameters in the frame after removing the influence of the first multi-pulse from the audio signal; , The output code of the parameter calculation circuit and the output code of the sound source calculation circuit Characterized by combination output.

The speech decoding apparatus according to the present invention inputs a discrete speech signal, extracts a spectrum parameter representing a spectrum envelope and a pitch parameter representing a pitch for each frame from the speech signal, and converts the speech signal of the frame into the pitch parameter. , And a first multi-pulse is obtained for one of the small sections as a sound source signal while performing pitch prediction using the spectrum parameter and the pitch parameter. After removing the influence of the multi-pulse from the audio signal, a second multi-pulse is determined without pitch prediction using the spectrum parameter in the frame in the frame, and the spectrum parameter, the pitch parameter, the first multi-pulse, and the Decoding the signal from the transmitting side transmitting the second multi-pulse A demultiplexer that separates and decodes a code representing the spectrum parameter, a code representing the pitch parameter, a code representing the first multi-pulse, and a code representing the second multi-pulse. A circuit, and divides a frame section into small sections corresponding to the decoded pitch parameter, generates the decoded first multi-pulse for one of the small sections, and reproduces a pitch using the pitch parameter; A sound source restoring circuit for restoring a sound source signal by generating and adding the decoded second multi-pulse to the signal whose pitch has been reproduced, and a synthesized speech signal using the restored sound source signal and the composited spectrum parameter. And a synthesis filter circuit for obtaining and outputting.

(Operation) In the speech encoding / decoding method and apparatus according to the present invention, a multipulse (first multipulse) obtained by pitch prediction of a sound source signal of a speech signal in a frame section and a pitch obtained without pitch prediction are obtained. It is characterized by using multi-pulses (second multi-pulses). Further, the first
In the multipulse calculation, the frame is divided into small sections (subframes) corresponding to the pitch cycle in advance in order to use the quasi-periodicity of each pitch of the multipulse sound source very efficiently and to greatly reduce the amount of calculation. The multi-pulse is divided by pitch prediction for only one section of the subframe. Then, after the signal is reproduced by the multi-pulse and the influence is removed, the multi-pulse is obtained without pitch prediction in the frame by the same method as in the document 1.

Hereinafter, the basic concept of the present method will be described with reference to FIG. FIG. 2 is a block diagram showing the operation of the present invention. An audio signal is input from the input terminal 100, and the audio signal is divided into frames of a predetermined time length (for example, 20 ms). From the audio signal of the frame, the LPC analysis unit 150 calculates a LPC coefficient of a predetermined order representing a spectral envelope to a known LPC coefficient.
Determined by PC analysis. As the LPC coefficient, other good parameters such as LSP, formant, and LPC cepstrum can be used in addition to the linear prediction coefficient. In addition, analysis methods other than LPC, for example, cepstrum, PSE, ARMA method and the like can also be used. In the following, description will be made on the assumption that a linear prediction coefficient is used. The pitch calculation circuit 200 calculates a pitch period M and a pitch coefficient (gain) from the audio signal of the frame.
Calculate b. A well-known autocorrelation method can be used for this. For calculating the pitch coefficient b (gain), a method using the value of the autocorrelation coefficient at the time delay M in the autocorrelation, or dividing the audio signal into small sections (subframes) for each pitch period M, A gradient value obtained when the rms value of the audio signal or the prediction residual signal in the subframe is approximated by a linear regression line can also be used. For the latter method, see Ono et al., “2.4Kbps pitch predictio
n multi-pulse speech coding "(proc.IEEE ICASSP88,
S4.9, 1988) can be referred to.

The operation of the pitch prediction multi-pulse calculation unit 250 and the multi-pulse calculation unit 270 will be described with reference to FIG. FIG. 3A shows an audio signal of a frame. Here, the frame length is set to 20 ms as an example. First, the pitch prediction multi-pulse calculation unit 250 divides a frame into small sections (subframes) using a pitch period M as shown in (b). Here, the subframe length is the same as the pitch period M. Next, the impulse response h _w (n) of the cascade connection filter of the pitch reproduction filter and the perceptually weighted spectrum envelope synthesis filter is obtained by the same method as in the above-mentioned document 2. Here, the transfer characteristics of the spectrum envelope synthesis filter and the pitch reproduction filter are expressed by the equations (1) and (2), respectively.

_{H p (z) = 1 /} (1-bz- M) (2) where the pitch reproduction filter order is set to 1.

The impulse responses of the equations (1) and (2) are represented by h _s (n) and h _p
Assuming that (n) and the impulse response of the auditory weighting filter is w (n), the impulse response h _w (n) is expressed by the following equation.

_{h w (n) = h s} (n) * h p (n) * w (n) (3) Further, the impulse response of the spectral envelope synthesis filter was subjected to perceptual weight moisture h _ws (n) is h _ws ( n) = h _s (n) * w (n) (4) Here, the symbol “*” represents convolution.

Next, the autocorrelation number R _hh (m) of the impulse response h _w (n), the perceptually weighted audio signal, the impulse response h _w (n), and the cross-correlation function Φ _hx
(M) is obtained. Next, the amplitude g _{i of} the predetermined number L (here, 4) of the multi-pulses is determined for only one predetermined section (for example, the section of FIG. 3B) of the sub-frame. determining a position m _i by the pitch prediction. FIG. 3 (c) shows a multi-pulse obtained. Next, through the pitch reproduction filter defined by the multi-pulse (2) obtained FIG. 3 (d Then, a pulse is reproduced in another sub-frame as shown in Fig. 7. Next, a reproduced signal x '(n) is obtained by driving the spectral envelope synthesis filter defined by the equation (1) using the reproduced pulse.

The subtractor 260 subtracts x '(n) from the audio signal x (n) to obtain e (n). Next, the multi-pulse calculator 270
Is the cross-correlation function between e (n) and the weighted impulse response of the spectral envelope synthesis filter for e (n), using the same method as in Reference 1, Using the autocorrelation function of the weighted impulse response, a predetermined number Q of multipulses in the frame is obtained. This is shown in FIG. In the figure, Q
Is set to 7.

Transmission information on the transmission side includes a spectrum parameter representing a spectrum envelope, a pitch period M of a pitch reproduction filter, a gain b, an amplitude and a position of L pitch prediction multi-pulses,
The amplitudes and positions of Q multipulses.

FIG. 1 shows an embodiment of the present invention.
A discrete audio signal x (n) is input from 0. In the spectrum / pitch parameter calculation circuit 520, the spectrum parameter of the spectrum envelope synthesis filter representing the spectrum envelope of the audio signal in the divided frame section (for example, 20 ms)
a _i is determined by a well-known LPC analysis method. Further, the coefficient b and the pitch period M of the pitch reproduction filter are obtained by a well-known autocorrelation method or the method described in the aforementioned reference 3.

The quantizer 525 performs quantization on the obtained spectral parameters, coefficients, and pitch period.
The method of quantization is described in Japanese Patent Application No. 59-272435 (Reference 4).
Scalar quantization or vector quantization as shown in FIG. For a specific method of vector quantization, see, for example, “Vecto
r quantization in speech coding "(Proc.IEEE, pp.155
1-1588, 1985) (Reference 5). The inverse quantizer 530 inversely quantizes using the result of the quantization and outputs the result.

The subtractor 535 subtracts the influence signal from the audio signal of the frame and outputs the result. The weighting circuit 540 performs perceptual weighting on the signal using the audio signal and the inversely quantized spectral parameter. The weighting method can be referred to the weighting circuit 200 of Reference 2.

The impulse response calculation circuit 550 calculates the impulse response h of the spectral envelope synthesis filter that has been subjected to the perceptual weighting using the inversely quantized spectral parameter a ^{′ ′} _i ▼.
_ws (n), and wherein (1), (2) cascaded consisting weights not a only the impulse response hw filters (n) and the a ▲ ^'
Calculation is performed using _i ▼, pitch period M ′, and coefficient b ′. The specific method can be referred to the impulse response calculation circuit of the above reference 2.

The autocorrelation function calculation circuit 560 calculates two types of autocorrelation functions for the two types of impulse responses, and outputs them to the sound source pulse calculation circuit 580 and the pulse calculation circuit 586, respectively. The calculation method of the autocorrelation function can be referred to the autocorrelation function calculation circuit 180 of the above-mentioned reference 2 or 4.

The cross-correlation function calculation circuit 570 calculates a cross-correlation function Φ _xh (m) between the perceptually weighted signal and the impulse response h _w (n).

The sound source pulse calculation circuit 580 first divides the frame into subframe sections as shown in FIG. 3B using the pitch period M 'obtained by dequantizing the frame. Then, for one predetermined subframe section (for example, the subframe in FIG. 3B), the amplitude g _{i of} the L multipulse trains is calculated using Φ _xh (m) and R _hh (m). determine the position m _i. For the method of calculating the pulse train, reference can be made to the sound source pulse calculation circuit of Reference 2.

The quantizer 585 quantizes the amplitude and position of the multi-pulse train and outputs a code. The specific method is described in the aforementioned reference 4.
And so on. This output is further dequantized, and the pitch reproduction filter 605 and the spectrum envelope synthesis filter 610
To obtain a synthesized speech signal x '(n) by the pitch prediction multi-pulse.

The subtracter 615 subtracts the synthesized voice signal x '(n) from the voice signal x (n) to generate a residual signal e (n).
Get.

The weighting circuit 600 performs perceptual weighting on the residual signal.

The cross-correlation function calculation circuit 603 calculates a cross-correlation function between the output of the weighting circuit 600 and the impulse response h _ws (n).

In the pulse calculation circuit 586, the amplitude and the position of a predetermined number of multi-pulses are obtained using the cross-correlation function and the auto-correlation function of the impulse response h _ws (n).

The digitizer 620 quantizes and outputs the amplitude and position of the multi-pulse, and inversely quantizes these and outputs the result to the synthesis filter 625.

 The combining filter 625 combines and outputs the residual signal.

An adder 627 adds the outputs of the synthesis filter 625 and the synthesis filter 610 to obtain a frame reproduction signal, and further obtains and outputs an influence signal for the next frame. The specific method can be referred to the above-mentioned reference 4.

The multiplexer 635 combines and outputs a code representing the amplitude and position of the multi-pulse train output from the quantizers 585 and 620, a spectrum parameter outputted from the parameter quantizer 525, a pitch period, and a code representing a coefficient.

On the other hand, on the receiving side, the demultiplexer 710 includes a code representing the amplitude and position of the pitch prediction multipulse (first multipulse), a code representing the amplitude and position of the multipulse (second multipulse), a spectrum parameter, The codes representing the pitch period and the coefficient are separated and output.

The first pulse decoder 720 decodes the amplitude and position of the pitch prediction multi-pulse. The second pulse decoder 725 is
The multi-pulse amplitude and position are decoded. Spectrum,
The pitch parameter decoder 750 includes an inverse quantizer 530 on the transmission side.
In the same manner as described above, the spectrum parameter, pitch period M 'and coefficient b' are decoded and output.

The first pulse generator 720 generates and outputs a sound source signal based on the first multi-pulse (pitch prediction multi-pulse) for a frame length. The second pulse generator 725 generates a sound source signal based on the second multi-pulse by a frame length.

The pitch reproduction filter 735 performs the same operation as the transmission-side pitch reproduction filter 605, and receives the output of the pulse generator 730, the decoded pitch period M 'and the coefficient b', and
The sound source signal whose pitch is reproduced by the frame is obtained and the adder 740
Output to

An adder 740 adds the sound source signal and the output signal of the second pulse generator 727 to obtain a frame drive sound source signal, and drives the synthesis filter circuit 760.

The synthesis filter circuit 760 obtains and outputs a synthesized speech waveform for each frame using the driving excitation signal and the decoded spectrum parameter.

The configuration described above is merely an embodiment of the present invention, and various modifications are possible.

As a method of calculating the multi-pulse, various well-known methods can be used in addition to the method shown in the above-mentioned document 1. This includes, for example, “A Study on
Pulse Search Algorithms for Multi-pulse Speech Co
der Realization "(IEEE JSAC, pp. 133-141, 1986) (Reference 6).

As a method of calculating the pitch period, in addition to the method shown in the above-described embodiment, for example, as shown in the following equation (3), a past sound source signal v (n), a pitch reproduction filter, and a spectrum envelope synthesis filter are used. A position M that minimizes the error power E between the signal reproduced by the above and the input audio signal x (n) of the current subframe can be searched for, and the coefficient b at that time can be obtained.

Here, h _s (n) indicates the impulse response of the spectrum synthesis filter, and w (n) indicates the impulse response of the auditory weighting circuit.

Further, a linear shift τ may be allowed in the pitch period M of the subframe. For specific instructions,
Reference 3 can be referred to. However, at this time, it is necessary to transmit τ in addition to the period M as pitch information.

Also, if the transmitting side synthesis filter 610 reproduces the weighted signal and subtracts it from the weighting circuit 540, the weighting circuit 600 can be omitted.

Also, the subtraction of the influence signal on the transmission side can be omitted. With such a configuration, the synthesis filter 625 and the adder 627 become unnecessary.

(Effect of the Invention) According to the present invention, a pulse having a strong periodicity for each pitch is represented very efficiently by obtaining a pulse in one subframe section by pitch prediction, and the correlation for each pitch is so strong. For non-existent pulses, multi-pulses are obtained without using pitch prediction, so that compared to the conventional method of obtaining all pulses using pitch prediction, sound quality is reduced in parts where periodicity is weak, such as vowel transitions and transitions. There is an effect that it can be greatly improved. Furthermore, since only some of the multi-pulses are obtained by pitch prediction, the amount of calculation required for searching for a pitch prediction multi-pulse can be significantly reduced, and the amount of calculation can be significantly reduced as compared with the conventional method. There is a big effect that.

[Brief description of the drawings]

FIG. 1 is a block diagram showing the configuration of an embodiment of a speech encoding / decoding method and apparatus according to the present invention, and FIG. 2 is a block diagram showing the operation of the present invention. FIG. 3 is a block diagram showing an example of a pulse train search. FIG. 4 is a block diagram showing an example of a conventional system. In the figure, 150: LPC analysis unit, 200: Pitch calculation unit, 2
50: Pitch prediction multi-pulse calculator, 270: Multi-pulse calculator, 520: Spectrum / Pitch parameter calculator, 525: Parameter quantizer, 530: Dequantizer, 535, 260, 615, 840 ... Subtractor, 540, 600, 850 ...
… Weighting circuit, 550 …… Impulse response calculation circuit, 560
…… Autocorrelation function calculation circuit, 570, 603… Cross correlation function calculation circuit, 585, 620 …… Quantizer, 627, 740 …… Adder, 586 …… Pulse calculation circuit, 605, 735 …… Pitch reproduction Filter, 610, 625, 760, 820 …… Synthetic filter, 635
…… Multiplexer, 710 …… Demultiplexer, 720…
... first pulse decoder, 725 ... second pulse decoder, 7
50 …… Spectrum / pitch parameter decoder, 730 ……
First pulse generator, 727... Second pulse generator.

Claims (3)

(57) [Claims] 1. A transmitting side receives a discrete voice signal, extracts a spectrum parameter representing a spectrum envelope and a pitch parameter representing a pitch for each frame from the voice signal, and converts a voice signal of the frame according to a pitch parameter. The first multi-pulse is obtained while performing pitch prediction using the spectrum parameter and the pitch parameter for one of the sub-sections as a sound source signal, and obtaining the first multi-pulse. After removing the influence of the pulse from the audio signal, a second multi-pulse was obtained without pitch prediction using the spectral parameter in the frame in the frame. The spectral parameter, the pitch parameter, the first multi-pulse, and the 2 multi-pulses are transmitted, and the receiving side receives the first multi-pulse. Scan and said pitch parameters and said second speech coding and decoding method and obtains the synthesized speech signal using the further the spectral parameters to restore the sound source signal using multi-pulse. 2. A parameter calculation circuit for extracting and encoding a spectrum parameter representing a spectrum envelope and a pitch parameter representing a pitch from an input discrete voice signal for each frame, and converting a voice signal of the frame into the pitch parameter. A dividing circuit that divides the signal into small sections corresponding to the first multi-pulse while performing pitch prediction for one of the small sections as a sound source signal using the spectrum parameter and the pitch parameter; A sound source calculation circuit for obtaining and encoding a second multipulse without pitch prediction using the spectral parameter in the frame after removing the influence of the first multipulse from the audio signal; and calculating the parameter. Combining the output code of the circuit and the output code of the sound source calculation circuit Speech coding apparatus characterized by force. 3. A discrete audio signal is input, and a spectrum parameter representing a spectrum envelope and a pitch parameter representing a pitch are extracted from the audio signal for each frame, and a speech signal of the frame is extracted according to the pitch parameter. The first multi-pulse is obtained while performing pitch prediction using the spectrum parameter and the pitch parameter for one of the sub-sections as a sound source signal. After removing the effects from the audio signal, a second multipulse is obtained without pitch prediction using the spectral parameters in the frame, and the spectral parameter, the pitch parameter, the first multipulse, and the second multipulse are determined. Speech decoding to decode the signal from the transmitter transmitting the pulse A demultiplexer circuit for separating and decoding a code representing the spectrum parameter, a code representing the pitch parameter, a code representing the first multipulse, and a code representing the second multipulse, and a frame. Dividing a section into small sections corresponding to the decoded pitch parameter, generating the decoded first multipulse for one of the small sections, reproducing the pitch using the pitch parameter, and reproducing the pitch A sound source restoring circuit for restoring a sound source signal by generating and adding the decoded second multi-pulse to the decoded signal, and obtaining and outputting a synthesized speech signal using the restored sound source signal and the decoded spectrum parameter. A speech decoding device comprising: a synthesis filter circuit.