JPH01261930A - Sound encoding/decoding system - Google Patents

Abstract

PURPOSE:To obtain satisfactory sound quality irrespective of voice and silent sound by changing the load coefficient of a noise shaping filter and a post noise shaping filter in accordance with the abutment condition of a prediction signal and encoding/decoding a digital input sound signal. CONSTITUTION:A load coefficient selector 66 evaluates the abutment condition of prediction by using digital input sound, short time prediction residual and long time prediction residual signals. Thus, a short time prediction parameter from an LPC parameter/short time prediction parameter converter and a pitch parameter from a decoder are adaptively loaded. They are set in short time prediction unit pole and zero filters 62 and 63, and long time prediction unit pole and zero filters 58 and 59. The result of a prescribed processing is transmitted to a subtractor as the output of the noise shaping filter 19. The transmission function F(z) is given by an expression I, and the filter 19 constitute the long time prediction unit pole and zero filters 58 and 59, and the short time prediction unit pole and zero filters 62 and 63 so that it satisfies the expression I.

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of Industrial Application) The present invention relates to an audio encoding/decoding method for encoding/decoding a digital input audio signal at a low bit rate.

(Prior art) In systems with severe frequency band restrictions and transmission power restrictions, such as digital maritime satellite communication systems and 5cpc digital business satellite communication systems, low bit rate and high quality encoding processing is required. There is a need for a voice encoding/decoding system that can obtain voice and has very little influence from transmission code errors.

Against this background, various audio encoding/decoding methods have already been proposed, and a typical method uses a predictor that calculates prediction coefficients for each frame and adapts residual signals by removing correlations between sample values. There is an adaptive predictive coding (APC) method in which coding is performed using a quantizer, and a multi-pulse driven linear predictive coding (MPEC) method in which an LPC synthesis filter is driven using a plurality of pulses as sound sources.

Here, an adaptive predictive coding method will be explained as an example of a conventional audio encoding/decoding method.

FIG. 1(a) is a block diagram of an encoder used in a conventional adaptive predictive coding method. First, the operation will be explained. A digital input speech signal Sj is input to an LPC analyzer 2 and a short-term predictor 6 via a coding input terminal 1. LPC
The analyzer 2 performs short-time spectrum analysis (hereinafter referred to as rLPC analysis) for each frame based on the digital input audio signal, and the LPC parameters obtained therefrom are used as L'P
encoded via C-parameter encoder 3 and multiplexed circuit 30
to the decoder on the receiving side. Also, LP'C
The output of the parameter encoder 3 is sent to the LPC parameter decoder 4.
The LPC parameter/short-term prediction parameter converter 5 obtains short-term prediction parameters from the output thereof. Then, this short-time prediction parameter is set in the short-time predictor 6, the noise shaping filter 19, and the short-time predictor 24 for local decoding.

The subtracter 11 subtracts the output of the short-time predictor 6 using this short-time prediction parameter from the digital input audio signal Sj, thereby removing the correlation between adjacent samples of the audio waveform and producing a short-time prediction residual signal ΔSj. get. This short-term prediction residual signal ΔSj is input to the pitch analyzer 7 and the long-term predictor 10. The pitch analyzer 7 performs pitch analysis for each frame based on the short-time prediction residual signal ΔSj, and the pitch period and pitch parameter obtained there are encoded via the pitch parameter encoder 8, and the multiplex circuit 30
to the decoder on the receiving side. - On the other hand, the pitch period and the pitch parameter are decoded via the pitch parameter decoder 9, the long-term predictor 10, the noise shaping filter 19,
It is set in the long-term predictor 23 for local decoding.

The subtracter 12 subtracts the output of the long-term predictor 1o using this pitch period and pitch parameter from the short-term prediction residual signal ΔSj, thereby removing the correlation of the repetitive waveform due to the pitch of the audio signal and ideally Obtain a whitened long-term prediction residual signal. The output of the noise shaping filter 19 is subtracted from this long-term prediction residual signal using a subtracter 17, and the final prediction residual signal is calculated by the adaptive quantizer 1.
6, the signal is quantized and encoded, and transmitted to the receiving side decoder via the multiplexing circuit 3o. Further, this encoded final prediction residual signal is decoded and dequantized via the dequantizer 18 and inputted to the subtracter 20 and the adder 21. The subtracter 20 obtains quantization noise by subtracting the final prediction residual signal, which is the input signal of the adaptive quantizer 16, from this quantized final prediction residual signal, and inputs this to the noise shaping filter 19. do.

In order to update the quantization step size for each subframe, the RMS value of the above-mentioned long-term prediction residual signal is calculated in the RMS calculation circuit 13, encoded in the RMS value encoder 14, and the output level is used as a reference. level, and also store nearby levels in the RMS value encoder 14. and,
The output signal of the RMS value encoder 14 is decoded via the RMS value decoder 15, and in particular, the quantized RMS value corresponding to this reference level is used as the reference RMS value, and a basic step size prepared in advance is set to this value. The step size of the adaptive quantizer 16 is determined by multiplying and summing . On the other hand, the output of the local decoding long-term predictor 23 is added to the quantized final prediction residual signal, which is the output signal of the inverse quantizer 18, via the adder 21. Furthermore, this is input to the long-term predictor 23 for local decoding, and the output of the short-term predictor 24 for local decoding is added via the adder 22, and this is input to the short-term predictor 24 for local decoding. do. The digital input audio signal Sj' locally decoded by this process
is obtained. The difference between this locally decoded digital input audio signal Sj' and the original digital input signal Sj is determined as an error signal via a subtracter 26. Between subframes, the power of this error signal is detected by the minimum error power detector 2.
Calculate by 7. Then, the same series of operations is performed for the total basic step size prepared in advance and the levels near the stored reference level, and the code that gives the minimum power among the error signal powers obtained above is executed. RM
The S level and basic step size are selected and transmitted via the multiplex circuit 30 to the decoder on the receiving side. Note that when encoding the step size, the step size encoder 2
9 is used.

FIG. 1(b) is a block diagram of a decoder used in the conventional adaptive predictive coding method.

In the decoder, the signal input through the decoder input terminal 32 is demultiplexed into a signal related to the final residual signal, a signal related to the RMS value and step size, and a signal related to the LPC parameter and pitch period/pitch parameter.
3 and an adaptive inverse quantizer 36, RM
S value decoder 35, step size decoder 34, LPC parameter decoder 38, and pitch parameter decoder 37
is input.

The RMS value decoder 35 is used to decode the RMS value, and this and the basic step size obtained via the basic step size decoder 34 are set in the adaptive inverse quantizer 36. Then, the signal regarding the received final prediction residual signal is dequantized using the adaptive dequantizer 36 to obtain a quantized final prediction residual signal. On the other hand, LPC parameter decoder 3
The short-time prediction parameters obtained through the LPC parameter/short-time prediction parameter converter 39 are decoded through the LPC parameter/short-time prediction parameter converter 39, and are then decoded through the short-time predictor 43, which is one of the predictors forming the synthesis filter, and the post-noise shaping filter 44. Further, the pitch period and pitch parameter decoded via the pitch parameter decoder 37 are set in the long-term predictor 42, which is the other predictor forming the synthesis filter.

The adder 40 adds the output of the long-term predictor 42 to the output of the adaptive inverse quantizer 35, and adds the output to the long-term predictor 42.
By adding the output of the short-time predictor 43 to this input via the adder 41, a reproduced audio signal is obtained. Then, this signal is input to the short-time predictor 43, and is also input to the post-noise shaping filter 44, where noise shaping is performed. Furthermore, the signal is sent to the level adjuster 4.
5, and the level is adjusted by comparing it with the output of the post-noise shaping filter 44.

Next, the configurations of the noise shaping filter 19 and the post noise shaping filter 44 used in the encoder and decoder will be described.

FIG. 2 is a block diagram of a noise shaping filter 19 in a conventional encoder. To explain this operation, first, the short-time prediction parameters obtained from the LPC parameter/short-time prediction parameter converter 5 are input to the short-time predictor 49, and the pitch period and pitch parameters obtained from the pitch parameter decoder 9 are input to the short-time predictor 49. is set in the long-term predictor 47. Next, the quantization noise is input from the subtracter 20 to the long-term predictor 47,
The output is subtracted from the quantization noise via a subtractor 48. Then, the output of the subtracter 48 is input to the short-time predictor 49, and the output of the subtracter 48 and the output of the long-term predictor 47 are input to the subtracter 5.
It is added via 0 and sent to the subtracter 17. The transfer function F'(z) of this noise shaping filter is expressed by the following equation.

F'(z) = rnlPl (z) + (l-rnlPI (z))
Ps (z/rsrns) (1) Here, Ps(z)
and PI(z) are the transfer functions of the short-time predictor 6 and the long-term predictor 10 found from equations (2) and (3) described later, rs is leakage, and rnl and rns are long/short-time predictors. is called the noise shaping weighting coefficient of 0≦rs, r
It is a value in the range of nl, rns≦1. In conventional noise shaping filters, constant values are used for rnl and rns.

The transfer function Ps(z) of the short-time predictor 6 is given by the following equation.

Ns i =] Here, ai is a short-time prediction parameter, and Ns is the number of taps of the short-time predictor. ai is LPC for each frame
It is calculated within the analyzer 2 and the LPC parameter/short-time prediction parameter converter 5, and changes adaptively for each frame in response to the fluctuation of the spectrum of the input signal. The transfer function of the long-term predictor 10 can also be defined using a similar equation. In particular, when a one-tap predictor is used, the transfer function PI(z) is given by the following equation.

PI(z) = blz−Pp (3) Here, b
l represents a pitch parameter, and Pp represents a pitch period. These bl and Pp are also calculated by the pitch analyzer 7 on a frame-by-frame basis, and change adaptively on a frame-by-frame basis in response to fluctuations in the periodicity of the waveform of the input signal.

FIGS. 3(a) and 3(b) are block diagrams of a post-noise shaping filter 44 in a conventional decoder. In the conventional method,
A short-time post-noise shaping filter loaded with the short-time prediction parameter of equation (2) is used.

First, the operation of a post-noise shaping filter as shown in FIG. 3(a), which is composed of only pole filters, will be explained.

The short-term prediction parameters obtained from the LPC parameter/short-term prediction parameter converter 39 are set in the short-term predictor 52. The reproduced audio signal from the adder 41 and the output of the short-term predictor 52 are added via the adder 51, and the result is input to the short-term predictor 52 and the level adjuster 45. The transfer function Fp'(z) of this post-noise shaping filter 44 including the level adjuster 45 is expressed as follows.

Here, GO is the gain control parameter, rps
is called a noise shaping weight coefficient and has a value in the range of 0≦rps≦1.

Next, the operation of the post-noise shaping filter shown in FIG. 3(b), which has a configuration in which a zero filter is further added to FIG. 3(a), will be explained. The short-time prediction parameters obtained from the LPC parameter/short-time prediction parameter converter 39 are set in the pole filter 54 and zero filter 55 of the short-time predictor. The reproduced audio signal from the adder 41 and the output of the pole filter 54 are added via the adder 51, and this is used as input to the pole filter 54 and the zero filter 55. The output of the zero filter 55 is subtracted from the output of the adder 53 via a subtracter 56, and this output is input to the level adjuster 45. Level adjuster 45
The transfer function F p of this post-noise shaping filter including
o' (z) is as shown in the following equation.

where GO is the gain control parameter, rps
z and rpsp are called zero/pole noise shaping weighting coefficients, and have values in the ranges of 0≦rpsz and rpsp≦1, respectively.

In conventional post-noise shaping filters, these rps,
A constant value is used for rpSZ, ml)Sl).

(Problems to be Solved by the Invention) The noise shaping filter 19 in the conventional encoder has a filter configuration based on a predictive filter, and it shapes the spectrum of quantization noise into a form close to the speech spectrum and masks it with speech. This improves the perceptual sound quality. In particular, it is effective in improving the audibility of quantization noise that exists in bands with few audio signal components. However, the spectrum of a speech signal changes over time and has various spectral characteristics such as voiced and unvoiced sounds, but conventional noise shaping filters do not examine these characteristics of the speech signal and can be used as predictive filters. A constant weight is applied to the coefficient.

Therefore, if a large weighting coefficient that is optimal for unvoiced sounds is used, the voiced sounds will become muffled or distorted. Furthermore, using a small weighting coefficient that is optimal for voiced sounds does not necessarily result in sufficient noise shaping for unvoiced sounds. For this reason, with the conventional method using a constant weighting coefficient, sufficient voice quality cannot be obtained for both voiced and unvoiced sounds.

Further, the post-noise shaping filter 44 in the conventional decoder is a synthesis filter (long-term predictor 42 and short-term predictor 43).
) is a filter configuration that uses only the same short-time predictor 43 used in The quality is improved by increasing the level difference from noise in the frequency domain so that the noise is not noticeable to the auditory sense. Conventional post-noise shaping filters also apply a constant weight to short-term prediction parameters without examining the spectral features of the audio signal mentioned above. Therefore, if a large weighting coefficient that is optimal for unvoiced sounds is used, distortions and licks will occur in voiced sounds. Furthermore, if a weighting coefficient that is optimal for voiced sounds is used, noise shaping will not necessarily be sufficient for unvoiced sounds. For this reason, the conventional method using a constant weighting coefficient has the drawback that sufficient voice quality cannot be achieved for both voiced and unvoiced sounds.

On the other hand, in MPEC, the explanation of which is omitted, on the transmitting side, an error weighting filter is used for the error signal between the input audio and the reproduced audio obtained from the synthesis filter, and the drive pulse is adjusted so that the output power is minimized. In this case as well, the short-term prediction parameters in the synthesis filter multiplied by a constant weighting coefficient are used as error weighting filters, so the above-mentioned noise shaping filter 19 ( Similar to the post-noise shaping filter 44), there is a problem in that sufficient voice quality cannot be obtained for both voiced and unvoiced sounds.

The first object of the present invention was to solve the above-mentioned problems of the prior art, and to provide a speech encoding/decoding method that can obtain good speech quality regardless of voiced or unvoiced sounds. It's about doing.

The second and third objects of the present invention are to provide a noise shaping filter and a post noise shaping filter for realizing a speech encoding/decoding method that can obtain good speech quality regardless of voiced or unvoiced sounds. It is in.

(Means for Solving the Problems) The first feature of the present invention is to encode/decode a digital input audio signal by changing the weighting coefficients of the noise shaping filter and the post-noise shaping filter according to the degree of accuracy of the prediction signal. The reason lies in the fact that it is structured in such a way that it becomes

The second feature of the present invention is that the noise shaping filter in the encoder includes a weighting coefficient selector 66 and a long-term predictor pole filter 58.
and zero filter 59. The short-time predictor is configured using a pole filter 62 and a zero filter 63.

A third feature of the present invention is that the post-noise shaping filters in the decoder include a weighting factor selector (75° 76), a long-term predictor pole filter 68, and a zero filter 69. Using short-term predictor pole filter 72 and zero filter 73 and adding the outputs of long-term predictor pole filter 68 and zero filter 69, while short-term predictor pole filter 72 and zero filter 7
The reason is that the configuration is such that the output of 3 is subtracted.

(Structure of the invention) The present invention will be explained in detail below using the drawings.

In addition, in the following explanation, the configurations of the noise shaping filter used in the encoder and the post noise shaping filter used in the decoder, which are the differences between the present invention and the conventional one, will be detailed, and other configurations will be explained in order to avoid duplication of explanation. abbreviated to

FIG. 4 is a block diagram of a noise shaping □ filter used in the encoder according to the present invention. The weighting factor selector 66 receives the digital input audio signal from the encoder input terminal (coder input) 1, the short-time predicted residual signal from the subtracter 11, and the long-term predicted residual signal from the subtracter 12. The short-term prediction parameters from the LPC parameter/short-time prediction parameter converter 5 and the pitch parameters from the pitch parameter decoder 9 are weighted adaptively, and each , short-time predictor pole filter 62
, the short-term predictive difficult character filter 63, the long-term predictor pole filter 58, and the long-term predictive difficult character filter 59. Quantization noise from subtractor 20 and long-term predictor pole filter 5
8 in an adder 57, and the added output is input to a long-term predictor pole filter 58 and a zero filter 59.

Also, from the output of the adder 57, a long-time prediction difficult character filter 59 is used.
A subtracter 60 subtracts the output of 62 and zero filter 63. Then, the output of the short-time prediction difficulty filter 63 is subtracted from the output of the adder 61 by a subtracter 64, this difference output is further subtracted from the quantization noise by a subtracter 65, and the result is used as the output of the noise shaping filter 19. It is sent to the subtractor 17. The transfer function F (z) of this noise shaping filter is given by the following equation.

The noise shaping filter 19 of the present invention includes a long-term predictor pole filter 58, a zero filter 59,
The short-time predictor pole filter 62 and zero filter 63 may be configured. For example, the positions of the long-term predictor pole filter 58 and zero filter 59 and the short-term predictor pole filter 62 and zero filter 63 may be reversed. Also,
Although FIG. 4 shows a configuration example in which the weighting coefficient selector 66 is shared by the long-term predictor pole filter 58 and zero filter 59, and the short-term predictor pole filter 62 and zero filter 63, it is possible to select separate weighting coefficients. A container may also be used.

In general, voiced sounds have clear spectral envelopes, especially nasal sounds and word endings, which are close to sine waves, so short-term spectral prediction is easy to predict, and also because these sounds have a clear pitch structure. , pitch prediction is easy to hit,
Quantization noise is also low. On the other hand, unvoiced sounds such as fricatives have a spectrum similar to random noise and do not have a clear pitch structure, so short-term and long-term predictions are difficult to predict and there is a lot of quantization noise. Therefore, it is necessary to perform noise shaping that matches the characteristics of the voice by observing the accuracy of the prediction. For example, if the power of the digital input audio signal is Sk,
If the power of the short-time prediction residual signal is Rk, and the power of the long-term prediction residual signal is Pk, then the power ratio Sk/Rk of the audio signal before and after the short-time prediction processing, which is the input/output ratio of the predictor, and the total By using the power ratio Sk/Pk of the audio signal before and after the prediction process, it is possible to check the accuracy of the prediction, that is, the spectral structure of the input audio. This makes it possible to perform noise shaping weakly on portions where these ratios are large, such as voiced sounds, and strongly on portions where these ratios are small, such as unvoiced sounds. The weighting factor selector 66 in FIG. 4 is an example in which the input/output ratio of this predictor is used as the accuracy of prediction. Specifically, Sk/Pk.

Threshold 5thl for each Sk/Rk,
5th2 is provided, and the noise shaping weight coefficients rns, , rnl of the short- and long-term predictors are switched as shown in the following equation.

Sk/Pk > 5thl or Sk/Rk > 5
th2: rns = rnthl, rnl =
rnth3 or Sk/Pk≦5thl and Sk/R
k≦5th2, :rns=rnth2, rn
l = rnth4 (7) However, 0 ≦rn
t old≦rnth2≦1゜0≦rnth3≦rnth4≦
1. Also, as for the accuracy of the prediction, the coder input 1. of the weighting coefficient selector 66 in FIG. Subtractor 11 and subtractor 12
It is also possible to use the LPC parameter (reflection coefficient) ki output from the LPC parameter decoder 4 instead of the input information. Generally, 1 kil is close to 1 because the prediction gain is high for voiced sounds, nasal sounds, word endings, etc. Furthermore, for unvoiced sounds such as fricatives, 1 kiL is close to 0 because the prediction gain is low. Therefore, the parameter G that determines the prediction gain can be found from the following equation.

i=1 This formula shows that when the parameter G is close to 0, the prediction gain is high;
Since a value close to 1 means that the prediction gain is low, it is possible to distinguish between voiced sounds with a small parameter G and loud overtones or unvoiced sounds. Specifically, a threshold Gthl is set for this parameter 6, and the noise shaping weight coefficients rns and rnl of the short- and long-term predictors are determined and switched as shown in the following equation.

G < Gthl: rns = rnth5,
rnl = rnth7Gthl≦G: rns =
rnth6, rnl = rnth8 (9
) However, 0≦Gthl≦1゜0≦rnth5≦rnth6≦1.

0≦rnth7≦rnth8≦1 In addition to this, it is not limited to the four thresholds mentioned above, and more thresholds are provided, and the range of possible values of the parameter G is divided into smaller regions to switch the load coefficient. It is also possible.

FIG. 5 is a block diagram of a post-noise shaping filter 44 used in a decoder according to the present invention. The short-term prediction device loading coefficient selector 76 evaluates the accuracy of the prediction from the LPC parameter output from the LPC parameter decoder 38, and calculates the short-time prediction parameter output from the LPC parameter/short-time prediction parameter converter 39. The prediction parameters are weighted adaptively and the short-time predictor pole filter 72 and zero filter 7
Set to 3. Furthermore, the long-term predictor weight coefficient selector 75 similarly evaluates the accuracy of the prediction from the pitch parameter output from the pitch parameter decoder 37, and adaptively weights the same pitch parameter. The polar filter 68 and the zero filter 69 are set.

The reproduced audio signal from the subtracter 41 and the output of the long-term predictor pole filter 68 are added by an adder 67, and the result is input to the long-term predictor pole filter 68 and zero filter 690.

From the output of this adder 67, a long-term predictive difficult character filter 69
An adder 70 adds the added output to the output of the short-time predictor pole filter 72 , and an adder 71 adds the added output to the output of the short-time predictor pole filter 72 and inputs the result to the short-time predictor pole filter 72 and zero filter 73 . The output of this adder 71 and the output of the short-time prediction difficulty filter 73 are subtracted by a subtracter 74, and the result is used as the output of the post-noise shaping filter 44 and the level adjuster 4
Send to 5. The transfer function G (z) of the post-noise shaping filter of the present invention including the level adjuster 45 is given by the following equation.

However, rpsp, rpsz, rplp, and rplz are the noise shaping weight coefficients of the short-time predictor-like/zero filter, respectively;
Represents the noise shaping weighting coefficient of the long-term predictor-like/zero filter. This short-time predictor has spectral characteristics that maintain the formant structure of the LPC spectrum by overlapping the poles of the polar filter and the zeros of the zero filter with a smaller weight on the spectrum. This results in a characteristic in which the high frequency formant is emphasized more than the spectral characteristic of a simple polar filter. On the other hand, the long-term predictor is
By placing the poles between zero and zero, the spectrum has spectral characteristics that emphasize the pitch component more. In this way, by inserting the short-time prediction difficulty filter 73 and the long-term prediction gain filter 69 and further using the adder 70, the formant components of speech, especially the high-frequency formant components and pitch components, can be further improved. To emphasize,
It is possible to reproduce crisp and clear sound (sound that is not muffled).

For the same reason as the noise shaping filter in the signal generator, it is weak against parts where predictions are correct, such as voiced sounds, and strong against parts where predictions are not correct, such as unvoiced sounds. It becomes possible to perform plastic surgery. Specifically, first,
In the short-time predictor of the post-noise shaping filter, the case where the LPC parameter ki is used as the spectrum envelope information, that is, the case where, for example, the parameter G of equation (8) is used as the prediction gain as described above, will be described.

Threshold Gth2 for parameter G,
Gth3 is used to switch rpsp and rpsz as shown in the following equation.

G < Gth2: rpsp = rpsthl
, rpsz = rpsth4, Gth2≦G<
G th3: rpsp = rpsth2.

rpsz = rpsth5 (11) Gt
h3≦G: rpsp = rpsth3, r
psz -rpsth6However, 0≦Gth2≦Gth
3≦1゜0≦rpsthl≦rpsth2≦rpsth
3≦1゜0≦rpsth4≦rpstti5≦rpsL
h6≦l By switching the weighting coefficients of the short-time predictor pole filter 72 and zero filter 73 in this manner, it is possible to set coefficients that match the speech spectrum. The same thing can be done for the long-term predictor, and the above equation can also be used. Here, in order to simplify the explanation, a one-tap filter will be used. The pitch parameter bt represents the degree of correlation between pitches with a value in the range 0<bl<1, and as bl approaches 1, the pitch structure becomes clear and the long-term prediction gain becomes large. Here, a threshold bth is provided for bl, and rplp and rplz are switched as shown in the following equation.

bl < bth : rplp = rplth2.
rplz −rplth4bth≦bl: rpl
p = rplthl, rplz = rplth3
(12) However, 0≦bth≦ 1゜0≦rplthl≦rpltth2≦ 1゜0≦rplt
h3≦rplth4≦1 Similarly, by switching the weighting coefficients of the long-term predictor pole filter 68 and zero filter 69, coefficients that match the speech spectrum can be set.

In addition, in Fig. 5, the load factor selector (75
, 76) are the long-term predictor pole filter 58 and zero filter 59, and the short-term predictor pole filter 62 and zero filter 63.
Although a configuration example using different load factor selectors has been shown, the same load factor selector may be used as in FIG. 4.

Next, as a specific example of parameters, simulation results using a 9.6 KbpS maximum likelihood quantization adaptive predictive coding method (APC-MLQ) will be shown below.

(a) A case where the input/output ratio of the predictor is used as the degree of hit of the predicted signal to the noise shaping filter of the encoder (parameters in equation 7).

Sk/Pk >40 or Sk/Rk >30
: rns = 0.2゜rnl = 0.2 or Sk/Pk≦40 and Sk/Rk≦30:rn
s = 0.5, rnl = 0.5 (b) When the LPC parameter is used as the hit condition of the predicted signal to the post-noise shaping filter of the decoder (11
parameters of the expression).

G <0.08: rpsp -0,25,rpsz
= 0.1.0.08≦G<0.4: rps
p = 0.6, rpsz =0.3.0.4≦G
: rpsp = 0.9, rpsz = 0
．． 30(C) A case where the pitch parameter is used as the degree of hit of the predicted signal to the post-noise shaping filter of the decoder (parameters in equation 12).

bl < 0.4: rplp = 0.62. rp
lz = 0.5.0.4≦bl : rplp =
0.35. rplz = 0.5 (Effect of the Invention) As described above, the present invention encodes and decodes a digital input audio signal by changing the weighting coefficients of the noise shaping filter and the post-noise shaping filter according to the degree of accuracy of the predicted signal. By optimizing the audio quality, good audio quality can be obtained regardless of whether the sound is voiced or unvoiced. Further, the present invention can be easily implemented by using the input/output ratio of the predictor, the LPC parameter, the pitch parameter, etc. as the accuracy of the predicted signal.

In addition, by configuring the noise shaping filter with a weighting coefficient selector 66, a long-term predictor pole filter 58 and zero filter 59, and a short-term predictor pole filter 62 and zero filter 63, noise shaping is performed more strongly than in the conventional structure. Therefore, it is possible to reduce quantization noise.

Additionally, the post-noise shaping filter uses a weighting factor selector (75
, 76), long-term predictor pole filter 68 and zero filter 69 . Short-time predictor pole filter 72 and zero filter 73
By using a configuration in which the outputs of the long-term predictor pole filter 68 and zero filter 69 are added, and the outputs of the short-term predictor pole filter 72 and zero filter 73 are subtracted, this structure is more advantageous than the conventional structure. Since noise shaping can be performed strongly, it is possible to reduce quantization noise and reproduce clear and well-defined speech. Therefore, the speech encoding/decoding method using the noise shaping filter and the post-noise shaping filter according to the present invention can realize a low bit rate, high efficiency speech encoding/decoding method, and its effects are It is extremely large.

[Brief explanation of the drawing]

Figures 1 (a) and (b) are block diagrams of conventional audio encoding/decoding systems and audio decoding systems, Figure 2 is a block diagram of a noise shaping filter used in a conventional encoder, and Figure 3 (a
) and (b) are block diagrams of post-noise shaping filters used in conventional decoders, FIG. 4 is block diagrams of noise shaping filters used in encoders of the present invention, and FIG. 5 is block diagrams of post-noise shaping filters used in encoders of the present invention. FIG. 2 is a configuration diagram of a post-noise shaping filter used in FIG. 1... Encoder input terminal, 2... LPC analyzer, 3
...LPC parameter encoder, 4,38...LPC
Parameter decoder, 5, 39...LPC parameter
Short-time prediction parameter converter, 6,24,43.49・
... Short-time predictor, 7... Pitch analyzer, 8... Pitch parameter encoder, 9.37... Pitch parameter decoder, 10, 23, 42.47... Long-term predictor, 11,12.17,20,26,48゜56.60,
64, 65.74...Subtractor, 13...RMS calculation circuit, 14...RMS value encoder, 15.35--R
MS value decoder, 16... adaptive quantizer, 18.36.
... Inverse quantizer, 19... Noise shaping filter, 21,
22゜40.41,50,51,53,57,61,6
7.70,71. ...Adder, 25...Local decoding signal terminal, 27...Minimum error power detector, 28...
- RMS value step size selector 9...Step size encoder, 30...Multiple circuit, 32...Decoder input terminal, 33...Polymorphism separation circuit, 34...Step size Decoder, 44... Post-noise shaping filter, 45... Level adjuster, 54, 62.72... Short-term predictor pole filter, 55, 63.73... Short-term prediction difficulty filter, 58 .68...Long-term predictor pole filter, 59.69...Long-term predictor zero filter, 6
6... Loading factor selector, 75... Loading factor selector for long time predictor, 76... Loading factor selector for short time predictor.

Claims (5)

[Claims] (1) On the transmitting side, there is a predictor that creates a predicted signal of the digital input audio signal, and a quantizer that calculates a residual signal between the predicted signal and the digital input audio signal and quantizes and encodes the residual signal. a dequantizer that decodes and dequantizes the output signal of the quantizer to obtain the residual signal; and a dequantizer that decodes and dequantizes the output signal of the quantizer to obtain the residual signal; a noise shaping filter that extracts it as quantization noise and shapes and reduces the quantization noise; and a receiving side decodes and dequantizes the encoded residual signal sent from the transmitting side; an inverse quantizer for
a synthesis filter for reproducing the digital input audio signal by adding the output signal of the inverse quantizer and the reproduced predicted signal; and a synthesis filter for reducing the quantization noise of the reproduced digital input audio signal. In the speech encoding/decoding method, the digital input speech signal is encoded/decoded by arranging a decoder having a post-noise shaping filter, and the noise-shaping filter and the post-noise shaping filter are arranged to encode/decode the digital input speech signal. A speech encoding/decoding method characterized by being configured to adaptively change coefficients depending on the degree of success. (2) The audio encoding/decoding method according to claim 1, wherein the input/output ratio of the predictor is determined and used as the accuracy of the predicted signal. (3) The audio encoding/decoding method according to claim 1, wherein an LPC parameter and a pitch parameter for creating the predicted signal are used as the accuracy of the predicted signal. (4) A residual signal that is the difference between the digital input audio signal and the predicted signal created by the predictor, and a decoded signal that is decoded by the inverse quantizer in the encoder after encoding the residual signal. In the noise shaping filter for reducing quantization noise, which is a difference, we use a short-time predictor pole filter and a short-time predictor zero filter that shape noise on the spectrum envelope, and a short-time predictor zero filter that shapes noise between pitches on the spectrum. a long-term predictor pole filter and a long-term predictor zero filter for determining the accuracy of the prediction signal; 1. A noise shaping filter comprising: a weighting coefficient selector for selecting a weighting coefficient of a zero filter. (5) After decoding and dequantizing the encoded residual signal sent from the transmitting side by an inverse quantizer, the output signal of the inverse quantizer and the predicted signal reproduced by the synthesis filter are combined. a short-time predictor that performs noise shaping on the spectral envelope of the reproduced audio signal in a post-noise shaping filter for reproducing the audio signal by adding the quantization noise of the reproduced audio signal. a pole filter and a short-time predictor zero filter; a long-term predictor pole filter and a long-term predictor zero filter that perform pitch-to-pitch noise shaping of the spectrum of the reproduced audio signal; and determine the accuracy of the predicted signal. and a weighting factor selector for selecting weighting factors of the short-term predictor pole filter, the short-term predictor zero filter, the long-term predictor pole filter, and the long-term predictor zero filter; The audio signal is processed by a long-term predictor pole filter and a long-term predictor zero filter whose optimal coefficients are set by a weighting coefficient selector, and then summed. A post-noise shaping filter characterized in that the reproduced audio signal is processed in the same manner by the filter and then subtracted.