CA1190323A - Adaptive predictive processing system - Google Patents
Adaptive predictive processing systemInfo
- Publication number
- CA1190323A CA1190323A CA000410713A CA410713A CA1190323A CA 1190323 A CA1190323 A CA 1190323A CA 000410713 A CA000410713 A CA 000410713A CA 410713 A CA410713 A CA 410713A CA 1190323 A CA1190323 A CA 1190323A
- Authority
- CA
- Canada
- Prior art keywords
- signal
- sub
- adaptive predictive
- band
- interval
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B1/00—Details of transmission systems, not covered by a single one of groups H04B3/00 - H04B13/00; Details of transmission systems not characterised by the medium used for transmission
- H04B1/66—Details of transmission systems, not covered by a single one of groups H04B3/00 - H04B13/00; Details of transmission systems not characterised by the medium used for transmission for reducing bandwidth of signals; for improving efficiency of transmission
- H04B1/667—Details of transmission systems, not covered by a single one of groups H04B3/00 - H04B13/00; Details of transmission systems not characterised by the medium used for transmission for reducing bandwidth of signals; for improving efficiency of transmission using a division in frequency subbands
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L19/00—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis
- G10L19/04—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis using predictive techniques
Landscapes
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Compression, Expansion, Code Conversion, And Decoders (AREA)
- Transmission Systems Not Characterized By The Medium Used For Transmission (AREA)
Abstract
ABSTRACT
An adaptive predictive coding system which splits a speech signal into a plurality of bands, subjecting each signal to adaptive predictive coding and adaptively changes quantization characteristics in the adaptive predictive coding so that quantization noise may become small. An average amplitude if the signal of each band is detected for each temporal sub-interval, and a quantization bit number and a quantization step size are adaptively determined from the average amplitude for each sub-interval, thereby permitting the reduction of the quantity of hardware used.
An adaptive predictive coding system which splits a speech signal into a plurality of bands, subjecting each signal to adaptive predictive coding and adaptively changes quantization characteristics in the adaptive predictive coding so that quantization noise may become small. An average amplitude if the signal of each band is detected for each temporal sub-interval, and a quantization bit number and a quantization step size are adaptively determined from the average amplitude for each sub-interval, thereby permitting the reduction of the quantity of hardware used.
Description
3~3 ADAP V PREDICTIVE PROCESSIMG SYST~M
BACKGROUND OF THE INVENTION
~he present invention relates to an adaptive predictive processing system which encodes, for instance, a speech signaL or musical signal, into predicti~e codes adaptively changing its coding characteristics i~ accordance with its property, or adaptively decodes the speech signal or musical signal from such encoded codes.
A conventional adap-tive predic-tive coding system is one that encodes the difference (a prediction residual) between a value linearly predic-ted from several previous sample vaLues of an input signal and a current input sample value and transmits the encoded difference. Various modifications can be effec-ted according to the arrangements of a predictor and a quan-tizer employed. For an unsteady inpu-t signal, such as a speech signal, it has been proposed to improve the signal-to-quantization noise ratio (SN ratio) by changing the predic-tion coefficient and the quantization ~tep size to comply with the statistic property of the signal. With the prior art adaptive predictive coding system, however, a bit rate above 32 Kb/s is needed for achieving quality equal to the toll quality and, at the bit rates below 16 Kb/s, the SN ratio is abruptly degraded and the quality is also markedly deteriorated by the quantization noise, Accordingly, the conventional system is not satisfactory.
An adaptive predictive coding system with adaptive bit allocation has been proposed as a system for improviny such performance degradation in the adaptive predictive coding at low bi-t ra-tes. According J.~ ~,
BACKGROUND OF THE INVENTION
~he present invention relates to an adaptive predictive processing system which encodes, for instance, a speech signaL or musical signal, into predicti~e codes adaptively changing its coding characteristics i~ accordance with its property, or adaptively decodes the speech signal or musical signal from such encoded codes.
A conventional adap-tive predic-tive coding system is one that encodes the difference (a prediction residual) between a value linearly predic-ted from several previous sample vaLues of an input signal and a current input sample value and transmits the encoded difference. Various modifications can be effec-ted according to the arrangements of a predictor and a quan-tizer employed. For an unsteady inpu-t signal, such as a speech signal, it has been proposed to improve the signal-to-quantization noise ratio (SN ratio) by changing the predic-tion coefficient and the quantization ~tep size to comply with the statistic property of the signal. With the prior art adaptive predictive coding system, however, a bit rate above 32 Kb/s is needed for achieving quality equal to the toll quality and, at the bit rates below 16 Kb/s, the SN ratio is abruptly degraded and the quality is also markedly deteriorated by the quantization noise, Accordingly, the conventional system is not satisfactory.
An adaptive predictive coding system with adaptive bit allocation has been proposed as a system for improviny such performance degradation in the adaptive predictive coding at low bi-t ra-tes. According J.~ ~,
2 -to this system, the quantization bit number and the quantiza-tion step size of a quantizer are adapted in accordance with non-uniformness in both of a frequency domain and a ~ime domain of residual power, and the prediction coefficient is computed from an autocorrelation coefficient of the input signal using a linear predictive analysis and adapted for each short time interval. Since this conventional system involves the computation of the autocorreLation and the residual. power for the adaptation of the quantization bit number, the quantization step size and the prediction coefficient, the word length representing these quantities has to be about twice the word length of the input signal in the case of implementing the system and a high-speed multiplier is required, resulting in the scale of hardware inevitably becoming large. Moreover, this system calls for transmission of the prediction coefficient to the receiving side separately of the residual signal.
SU~MARY OF THE INVENTION
It is therefore an object of the present invention to provide an adaptive predicti.ve coding system which permits transmission of a high quality speech signal even at relatively low bit rates in the range of 8 to 16 Kb/s and allows real~time processing with relatively small scale hardware.
Another object of the present invention is to provide an adaptive predictive coding system which does not necessitate the transmission of the prediction coefficient to the receiving side and hence is high in coding efficiency by that and highly stable.
Yet another object of the present invention is to provide an adaptive predictive decoding system which performs adaptive predictive decoding of a coc~e obtained by adaptation processing of a predictor and a quantizer.
According to the present invention, an input signal, such as a speech signal, is split into a plurality of frequency bands and the split signals are subjected to adaptive predictive coding for each band. ~t the same time an average amplitude of pseudo-prediction residual in each sub interval obtained by dividing the time axis for each band is detected and the quantization bit number and the quantization step size of a ~uantizer in the predictive coding for each corresponding band are adaptively changed in inter-band and time directions, thereby reducing a .quantization error. In addition, since the adaptive change of the quantization bit number and the quantization step size is carried out by detecting the average amplitude of the pseudo-prediction residual corresponding to the prediction residual in the predictive coding in each band, the word length for computation processing therefor is reduced, making it possible to reduce the scale of hardware by that. Furthermore, each predictive coder is formed by a lattice type digital filter. According to an embodiment of the present invention, a PARCOR coefficient is successively estimated from a locally decoded signal and, by forming a predictive filter for decoding in the same manner as the predictive code, the PARCOR coefficient need not be transmitted to the receiving side.
In accordance with an aspect of the invention there is provided an adaptive predictive processing system comprising splitting means for splitting an input signal into a plurality of frequency bands; sub-band translating means for translating into a sub-band signal the signal of each of the frequency bands split by the splitting means;
adaptive predictive coding means for subjecting each sub-band signal to adaptive predictive coding through utilization of a lattice filter in a predictor;
- 3a -pseudo-residual signal generating means for generating, for each sub-band signal, a pseudo-residual signal corresponding to a residual signal in the adaptive predictive coding means; sub-interval setting means for dividing the pitch period of the input signal at equal intervals into a plurality of sub-intervals and setting the sub-intervals with respect to an analysis frame;
amplitude detecting means for detecting an average amplitude of the pseudo-residual signal of each sub-band signal for each sub-interval; and quantization step size and quantiziation bit number allocation means for adaptively computing a quantization step size and a quantization bit number for a prediction residual signal in the adaptive predictive coding means from the average amplitude of the corresponding pseudo-residual signal for each sub-interval.
BRIEF DESCRIPTION OF THE DRAWINGS
Fig. l is a block diagram showing a conventional adaptive predictive coding system;
FigO 2 is a block diagram showing the arrangement of a predictive coefficient adaptation circuit 122 used ~ A -in Flg, 1;
Fig. 3 is a block diagram illustrating, by way of example, the arrangemen-t of a coder of the adap-tLve predictive processing system oE the present invention;
Fig, 4 is a diagram showing a specific example of adaptive predic-tive coders 8, 9 and 10 used in Fig. 3;
FigO 5 is a diagram showing a specific example of a PARCOR analyzer employed in Fig, 4;
Fig. 6 is a diagram showing a specific example of an inverse filter 12 utili2ed in Fig. 3;
E~ig, 7 is a cliagram explanatory of a me-thod for setting a sub-interval;
Fig, 8 is a block diagram illustrating, by way of example, a decoder of the adaptive predictive processing sys-tem of the present invention;
Yig, 9 is a diagram showing a specific example of predictive decoders 63, 64 and 65 used in FigO 8;
Fig, 10 is a block diagram illustrating another example of the coder in the adaptive predictive processing system of the present invention;
Fig. 11 is a flowchart showing an example of processing for obtaining Td;
Fig, 12 is a flowchart showing an example of processing for computing uij;
Fig, 13 is a flowchart showing a portion of processing for computing Ri ;
Fig, 14 is a flowchart showing an example of processing Eor obtaining an integer IRij of Rij; and Fig. 15 is a block diagram illustrating an embodimen-t of the presen-t invention as being applied to a multi-channel sys-tem.
-- 5 ~
DESC:RIPTION OF THE PREFERRE..D EMBODIMENTS
To facilitate a be-tter understanding of the presen-t invention, a brief description will be given first, with reference to Fig. 1, of the aforementioned conventional adaptive predictive coding sys-tem of the -type adaptively changing the quantization charac-teris^tics. A sampled i.nput signal of a speech in digital form, which is supplied from an input terminal 1, is split b~ sub-band splitting circuit 101, for instance, in-to three frequency bands and, at the same time, translated into sub-band signals. The sub band signals are subjected to adaptive predictive coding by adaptive predictive coders 103, 109 and 110, and the coded outputs of their residual si.gnals are provided to a code combiner 121. The adaptive predictive coders 103, 109 and 110 are each what is caLled a forward type, and recursive filters are used in predictors. From the sub-band signals are compu-ted, by predictive coefficient adaptation circuits 122, 123 and 124, predic-tion coefficients~ by which are adaptively modified the constants of filters of the predictors in the adaptive predictive coders 108, 109 and 110. The predictive coefficient adaptation circuit 122 is constructed as shown in Fig 2. The autocorrel.ation coefficients of the sub-band signal are computed by a correlator 127, and simultaneous linear equations using the computation result as a variable are solved by a predictive coefficient computation circuit 128 to obtain the prediction coefficient. The predictive coefficient adaptation circuits 123 and 124 are also identical in construction with the predictive coefficient adaptation circuit 122. The prediction coefficients ob-tained by the predictive coefficient adaptation circuits 122, 123 and 124 are supplied to the code combiner l21l too.
In Fig. 1, from the sub-band signals are derived by an inverse fil.-ter 112 pseudo residual. slgnals respectively corresponding -to the residual signals in the adaptive predictive coders 108 to 110. The pseudo-residual signals thus obtained are squared by a square computatlon circuit 129 to obtain power signals, which are averaged by an averager 114 to provide average powers of the pseudo-residual signals. The average powers of the pseudo-residual signals are supplied to an inter-sub-band bit allocation circuit l31 in a quantization bit number adaptation circuit 115. In the inter-sub-band bit allocation circuit 131 the quantization bit number allocation to the :Erequency bands is effected based on the rate of power among -the three pseudo-residual signals and then, in a temporal bit alloca-tion circuit 132, a quantization bit allocation adaptively takes place in accordance with the temporal localization of the powers of -the pseudo-residual signals. As a time interval for changing the bit aLlocation with -time, the pitch of the input speech signal is detected and the pitch period is divided at equal intervals into sub-intervals, whi.ch are set with respect to an analysis frame. The se-tting of the sub-intervals is carried out by a sub-interval se-tting circuit 133. For each sub~interval the average power of the pseudo-residual signal is obtained and the quantization bit allocation is performed.
The average powers of the pseudo-residual signals are subjected to square root computation by a square root computation ci.rcuit 134, and the thus computed square roots and the quantization bi-t numbers de-termined by the quantization bit number allocation circuit 115 are applied to a quantiza-tion step size adaptation circui-t 116, in which the quantization step size for each band is adaptively cletermined for each sub-interval. The quantization bit numbers and -the quan-tization step sizes -thus determined are se-t in quantizers in the adaptive predictive coders 108 -to 11 a correspondiny to the sub-bands, respectively.
The quantization bi-t numbers and the quantization s-tep sizes are de-termined so tha-t quantization error power may be small. Data indica-ting tne average powers of the pseudo-residual signals from -the averager 114, the pitch period from the sub-interval setting circuit 133 and -the positions of the sub-intervals relative to the analysis frame are provided to the code combiner 121. The code combiner 121 encodes the da-ta as a whole and derives the encoded output at an output terminal 18.
Accordiny -to the prior art adaptive predictive coding system described above, a quality substantially equal to PCM (six bits) of 48 Kb/s can be ob-tained at a bit ra-te of 16 Kb/s. Since -the predictive adaptation processing and the quantization adapta-tion processing take place based on power, however, this system has the defect that the quantity of hardware needed is large. Fur-thermore, since the adaptive predictive coders 108 to 110 are the forward type, the prediction coeEficients have -to be -transmitted, resulting in -the coding efficiency being lowered by tha-tO
Incidentally, in the case of employing recursive filters in the predictors and arranging the adaptive predictive coders in a backward type, the condition for stabiliza-tion is not dependent on the prediction coefficien-ts, so that when the prediction coefficients are adapted, no stabili-ty can be ensured.
Fig. 3 illustrates an embodiment o~ the adaptive predictive coding system of the present invention. A
speech-sampled input signal in digital form from the inpu-t -terminal 1 is spli-t by band-pass filters 2, 3 and 4 into a plurality oE frequency bands. Signals of -these divided ~ 8 ~
Erequency bands are re-sampled by samplers 5, 6 and 7 lnto sub-band signals~ which are subjected -to adap-tive predictive encoding by adap-tive predic-tive coders 8, 9 and 10 to derive -therefrom residual signals.
The adaptive predic-tive coders 8, 9 and 10 are each composed of a predictor (Ps) 19 based on -the correlation be-tween adjacent sample values and a predictor (Pp) 20 based on the correlation between sample values spaced one pi-tch period apart, as shown in Fig. 4. A
prediction signal is produced by obtaining the sum of prediction values from the both predictors 19 and 20 through the use of an adder 21. A prediction residual signal is produced by obtaining the difEerence between the input signal from a terminal 22a and the prediction siynal from the adder through the use of a subtractor 22. The prediction residual signal thus obtained is quantized by a quantiæer 23 and applied via a terminal 23a to a code combiner 17 shown in Fig. 3, and the encoded output in the code combiner 17 is provided via a terminal 18 on a transmission line.
The output from the quantizer 23 is provided to an inverse quantizer 24, too, as depicted in Fig, 4, wherein it is decoded. The decoded output from the inverse quantizer 24 is added by an adder 25 to the prediction value from the predictor 19, and the added output from the adder 25 is fecl back as a local decoded signal to the predictor 19. E'urther, the output from the adder 25 is added by an adder 26 to the prediction value from the predictor 20 and the added output is fed back to the predicto~ 20.
The predictor 19 is shown to be made up of P
stages of lattice digital filters, each stage being formed by one unit time lag element 27, two multipliers 28 and two adders in a manner to compu-te the following equation.
In-ternal signals of an ith (i = 1, 2, ... P) stage fil-ter, efi~t) and ebi(-t), are ob-tained by the following equa-tion from internal signals of an (i ~ 1)th stage filter, efi 1(t) and bi-1( ) (t) = efi_1(t) + Ki(t)ebi-1( ) e (-t) = Ki(t)efi_1(t ~ ebi~1( i 1, 2, . P, eb0(t) f0( where Ki(t) (i = 1, 2, ... P) is a PARCOR coei-f:icientt which is the multiplica-tor of the two multipliers 28r and ef0(t) is the input signal to the predictor 19. A prediction signal which is provided from an adder 30 is given by the following equdtion:
xS(t) = - ~1Ki(t)ebi-1( ) On the other hand, the predictor 20 is composed oE a singLe stage of lattice digital filter, in which a time lag elemen-t 34 represents a time lag dependent on the pitch period and the sampling period in each band and is implemented by an interpolating digital filter since the time lag usually assumes a value which is a non~integral multiple of the samp]ing interval of the input signal. The prediction siynal of the predictor 20, which is provided from an adder 32 is given by the following equa-tion:
xp(t) = - h(t)x(t - d) where h(t) is a PARCOR coefficient, x(t) is a local decoded signal which is -the input to the predic-tor 20 and d is the time lag o-f the time lag element 34. The PARCOR
coefficierl-ts lii(-t) and h(t) are obtained. b~ a PARCOR
analyzer 35 frorn the in-ternal signals efi(t) and ebi(t) for each unit -timeO These predictors 19 and 20 are well-kno~n in the art.
The PARCOR coefficients are given by the following equations:
Ki(t+1 ) = tanh~.en I fi(t) - ebi(t) ¦efi(t) + ebi(t) ___ _ ~ ¦x(-t) - x(t-d)¦
h(-t+1) = tanh ~Qn ¦x(t) ~ .x(t-d)¦
where (¦ ¦) deno-tes a time average. The above equa-tions can be realized by the PARCOR analyzer 35 as shown in Fig, 5, for instance, The signals efi(t) and ebi(t) are applied to adders 36 and 37, in which efi(t) - ebi(t) and efi(t) ebi(t) are ob-tained~ and these outputs are provided to absolute value computation circuits 38 and 39, wherein their absolute values are obtainedO The absolute values are smoothed by smoothing low pass filters 40 and 41, and logarithmlc values of the smoothed ou-tputs are obtained by logarithmic value computation circuits 42 and 43. The logari-thmic values -thus obtained are applied to an adder 44 to detect the difference therebetween. The difference output thus detected is provided to a hyperbolic tanyent (tanh x) compu-tation circui-t 45 to ob-tain a hyperbolic tangent, providing the PARCOR coefficient. The abovesaid absolute value computa-tion circuits 38 and 39 may also be replaced with squarers, The logarithmic value compu-tation circuits 42 and 43 and the hyperbolic tangent computation circui-t 45 can each be so arranged as -to read out clata storecl in a read only memory.
The quan-tization bit number and -the quantization step size of the quan-tizer 23 in each of the adaptive predictive coders 8, 9 and 10 in Fig. 2 are adapted in accordance with the mean amplitude of the pseudo-residual signal in -the present invention. The pseudo-residual siynal corresponds to the residual signal provided from the subtractor 22 (Fig. 4) in the adaptive predictive coder, and the mean ampli-tude oE the pseudo residual signal has to be compu-ted prior to the activation of the adaptive predictive coder. ~s shown in Fig, 3, from the sub-band signal of the lowest band, for example, the output from the sampler 5 is detected by a pitch period de-tector 11 the pi-tch period Tp of the signal, The detection of the pitch period can be effected in a known manner. That is, an average of the absolute value of the difference between sub-band sample values in the analysis frame is obtalned by ~_~
p(~ (t) - x(t+~)¦
M - T t=0 where M is the number of samples in the frame, and the pitch period is obtained as such a value of ~ that minimizes the average p~l) within the range cf ~min ~ T < ~maX~ By the pltch period Tp thus detec-ted, the delay time of the delay element 34 in Fig. 4 is controlled, In an inverse filter circuit 12 r f~m the sub-band signals of the respective bands are ob-tained theLr pseudo-prediction residual signals, The inverse filter 12 is one tha-t corresponds to the filter used for the predictors in the adaptive predictive coders 8, 9 and 10, and it is made up of, for instance, a cascade connection - 12 ~-~ .'3~
of an inverse filter 46 based on a pi.tch predic-tion and an inverse fi.l-ter ~7 based on an adjacent correlation as shown in Fig. 6; namely, i-t is formed by -the same lattice digital filters as those employed for -the adaptive predictive coder depicted in Fig, 4. Time delay elements 48 and 52 and PARCOR analyzers 51 are identical with the time delay elements 34 and 27 and the PARCOR analyzers 35 used i,n the adaptive predic-tive coder shown in Fig. 4.
Nex-t, that one of the outputs from the inverse fil,ter 12 which corresponds -to t:he sub-band signal of t,he lowest frequency band is provided to a sub-in-terval set.ting circuit 13, wherein the position of the sub-interval is detected based on the temporal localiza-tion of -the amplitude of its pseuclo-residual signal. As shown in Fig. 7, the pi-tch period Tp is divided at equal intervals into a plurality of sub-intervals (four sub-intervals i.n Fig. 7) so that they may repeat periodically. The positions of the sub-intervals are defined by the time l.ength Td from the beginning of the analysis frame -to the beginning 11 of a first sub-interval j = 11, 12 and 13. The time length Td is set so -that the mean amplitude in the first sub-interval may become maximum. Tha-t is to say, the mean amplitude is obtained by u(Td) ~ ~ ~T Ir1( )I
where T is the first sub-interval (11, 12, 13) when the time interval is set to a cer-tain value, M is the number of samples in the first sub-interval and r1(t) i.s the residual signal, and then Td which maximizes u(Td) is obta,ined in the range of 0 ~ Td < Tp, In this way, the sub-intervals are set.
The arithmetic processing for setting the temporal ~6~ 3 position I`d can be carried ou-t following a flowchart shown in Fig. 1l, for instance. In Fig~ 11, in step S1 the mean ampli-tucle u* of -the pseudo~residual. signal of the sub-interval i.s set to _~ and ~ is se-t as a variahle of Td to 0 and, in step S2, the mean amplitude of -the pseudo-residual signal of t~e sub-interval is set to 0 and the number of sample points, M is set to 0 and, further, a saMple point t is set to 0. In step S3 i-t is checked whether the sample point t lies within the sub-intervals in which the mean amplitude is to obtain, That is, T(t) represents the following equation in which 1 is added -to an in-tegral value obtained by dividing by the length of one sub-interval, TQ = Tp/]., the remainder of a division of t by the pitch period Tp:
~'(t) = [~ P-~] ~ 1 where [ ] is a Gaussian symbol and indicates -the largest integral value in [ ]. In the case of YES in step S3, the pseudo-residual. signal ¦r1(t)¦ of that sub-interval is added to u and the number of sample points, M, is incremented by 1 in step S~, which is ollowed by step S5. In the case of NO in step S3, the operation proceeds to step S5, in which the sample point t is incremented by 1, followed by checking in step S6 whe-ther the sample point t has become a final sample point a-t the end of the analysis frame ]ength Tf. In the case of NO in step S6, the operation goes back to step S3, whereas, in -the case of YES in step S6, the operation proceeds to step S7 in which the average amplitude of the pseudo-residual signal in the same sub-interval within the analysis frame is obtained. If it is decided in the next step S8 that the average amplitude u is larger than u*, -the average ampli-tude u is set as u* and ~ is set as Td in st.ep S9, and the operatlon proceeds to step S10.
Also in -the case of NO in -the step S8, the operation proceeds to step S10, In s-tep S10 the ~ is incremen-ted by 1, which is foLlowed by step S11, in which it is checked whether the ~ has become the pitch period Tp, If so, the operation comes -to an end and the Td :Ln step Sg at that time becomes a value representing the posi-tion oE ~he sub-interval rela-tive to the analysis Erame. If not, the operation returns to step S2 and the above-described operation takes place again to obtain the -time length Td which ma.xlmizes the average amplitude of the pseudo-residual signal of -the same sub-interval, In an average amplitude computation circuit 14 shown in Fig. 3, an average amplitude of the pseudo-residual signal in each sub-interval for each band is obta:ined by the following equation:
ij Ml~ t~Tij j - 1, 2, L
where Tij and Mij indicate a jth sub-interval in an ith band and the number of samples contained in the sub-interval, respectively, N is the number of bands into which the input signal is split (N = 3 in Fig. 3) and L is the number of periods into which -the pitch period is split (L
= 4 in Fig. 7).
The arithmetic processing -for obtaining the average amplitude of the pseudo-residual signal for each sub-interval is performed, for example as shown in Fig.
12. In step S12 operation starts with initialization of i = 1 and, in s-tep S13~ uij = 0 and Mij = 0 are initiali~ed for each j = 1, 2, ... L~ followed by setting of the sample poin-t t to 0 in step S14. In step S15 -the sub~interval ~ 15 -?a j is obtained by the equa-tion of-. T(t) from T(-t - Td) and, for the sub-interval. j, u j -~ ¦ri(t1)¦ is compu-ted, that is, -the pseudo-residual signals are accumulated and, furtner, Mij is incremented by 1. In the nex-t step S16 t is incremen-ted by 1 and in step S17 it ls checked whether t has become the final sample point Tfi. If so, the opera-tion proceeds -to step S18 and if not, -the operation returns -to s-tep S15. In step S18 the accumulated uij is divided by the number of sample poin-ts in the same sub-interval, Mij, -to obtain the average amp]itude Eor each j = 1,2, ... ~" In step Slg i i.s incremented by 1 and in step S20 it i.s checked whether i is larger -than N If not, the opera-tion returns to step S12 and if so, the operation comes to an end.
A quantization bit allocation circui-t 15 in Fig.
SU~MARY OF THE INVENTION
It is therefore an object of the present invention to provide an adaptive predicti.ve coding system which permits transmission of a high quality speech signal even at relatively low bit rates in the range of 8 to 16 Kb/s and allows real~time processing with relatively small scale hardware.
Another object of the present invention is to provide an adaptive predictive coding system which does not necessitate the transmission of the prediction coefficient to the receiving side and hence is high in coding efficiency by that and highly stable.
Yet another object of the present invention is to provide an adaptive predictive decoding system which performs adaptive predictive decoding of a coc~e obtained by adaptation processing of a predictor and a quantizer.
According to the present invention, an input signal, such as a speech signal, is split into a plurality of frequency bands and the split signals are subjected to adaptive predictive coding for each band. ~t the same time an average amplitude of pseudo-prediction residual in each sub interval obtained by dividing the time axis for each band is detected and the quantization bit number and the quantization step size of a ~uantizer in the predictive coding for each corresponding band are adaptively changed in inter-band and time directions, thereby reducing a .quantization error. In addition, since the adaptive change of the quantization bit number and the quantization step size is carried out by detecting the average amplitude of the pseudo-prediction residual corresponding to the prediction residual in the predictive coding in each band, the word length for computation processing therefor is reduced, making it possible to reduce the scale of hardware by that. Furthermore, each predictive coder is formed by a lattice type digital filter. According to an embodiment of the present invention, a PARCOR coefficient is successively estimated from a locally decoded signal and, by forming a predictive filter for decoding in the same manner as the predictive code, the PARCOR coefficient need not be transmitted to the receiving side.
In accordance with an aspect of the invention there is provided an adaptive predictive processing system comprising splitting means for splitting an input signal into a plurality of frequency bands; sub-band translating means for translating into a sub-band signal the signal of each of the frequency bands split by the splitting means;
adaptive predictive coding means for subjecting each sub-band signal to adaptive predictive coding through utilization of a lattice filter in a predictor;
- 3a -pseudo-residual signal generating means for generating, for each sub-band signal, a pseudo-residual signal corresponding to a residual signal in the adaptive predictive coding means; sub-interval setting means for dividing the pitch period of the input signal at equal intervals into a plurality of sub-intervals and setting the sub-intervals with respect to an analysis frame;
amplitude detecting means for detecting an average amplitude of the pseudo-residual signal of each sub-band signal for each sub-interval; and quantization step size and quantiziation bit number allocation means for adaptively computing a quantization step size and a quantization bit number for a prediction residual signal in the adaptive predictive coding means from the average amplitude of the corresponding pseudo-residual signal for each sub-interval.
BRIEF DESCRIPTION OF THE DRAWINGS
Fig. l is a block diagram showing a conventional adaptive predictive coding system;
FigO 2 is a block diagram showing the arrangement of a predictive coefficient adaptation circuit 122 used ~ A -in Flg, 1;
Fig. 3 is a block diagram illustrating, by way of example, the arrangemen-t of a coder of the adap-tLve predictive processing system oE the present invention;
Fig, 4 is a diagram showing a specific example of adaptive predic-tive coders 8, 9 and 10 used in Fig. 3;
FigO 5 is a diagram showing a specific example of a PARCOR analyzer employed in Fig, 4;
Fig. 6 is a diagram showing a specific example of an inverse filter 12 utili2ed in Fig. 3;
E~ig, 7 is a cliagram explanatory of a me-thod for setting a sub-interval;
Fig, 8 is a block diagram illustrating, by way of example, a decoder of the adaptive predictive processing sys-tem of the present invention;
Yig, 9 is a diagram showing a specific example of predictive decoders 63, 64 and 65 used in FigO 8;
Fig, 10 is a block diagram illustrating another example of the coder in the adaptive predictive processing system of the present invention;
Fig. 11 is a flowchart showing an example of processing for obtaining Td;
Fig, 12 is a flowchart showing an example of processing for computing uij;
Fig, 13 is a flowchart showing a portion of processing for computing Ri ;
Fig, 14 is a flowchart showing an example of processing Eor obtaining an integer IRij of Rij; and Fig. 15 is a block diagram illustrating an embodimen-t of the presen-t invention as being applied to a multi-channel sys-tem.
-- 5 ~
DESC:RIPTION OF THE PREFERRE..D EMBODIMENTS
To facilitate a be-tter understanding of the presen-t invention, a brief description will be given first, with reference to Fig. 1, of the aforementioned conventional adaptive predictive coding sys-tem of the -type adaptively changing the quantization charac-teris^tics. A sampled i.nput signal of a speech in digital form, which is supplied from an input terminal 1, is split b~ sub-band splitting circuit 101, for instance, in-to three frequency bands and, at the same time, translated into sub-band signals. The sub band signals are subjected to adaptive predictive coding by adaptive predictive coders 103, 109 and 110, and the coded outputs of their residual si.gnals are provided to a code combiner 121. The adaptive predictive coders 103, 109 and 110 are each what is caLled a forward type, and recursive filters are used in predictors. From the sub-band signals are compu-ted, by predictive coefficient adaptation circuits 122, 123 and 124, predic-tion coefficients~ by which are adaptively modified the constants of filters of the predictors in the adaptive predictive coders 108, 109 and 110. The predictive coefficient adaptation circuit 122 is constructed as shown in Fig 2. The autocorrel.ation coefficients of the sub-band signal are computed by a correlator 127, and simultaneous linear equations using the computation result as a variable are solved by a predictive coefficient computation circuit 128 to obtain the prediction coefficient. The predictive coefficient adaptation circuits 123 and 124 are also identical in construction with the predictive coefficient adaptation circuit 122. The prediction coefficients ob-tained by the predictive coefficient adaptation circuits 122, 123 and 124 are supplied to the code combiner l21l too.
In Fig. 1, from the sub-band signals are derived by an inverse fil.-ter 112 pseudo residual. slgnals respectively corresponding -to the residual signals in the adaptive predictive coders 108 to 110. The pseudo-residual signals thus obtained are squared by a square computatlon circuit 129 to obtain power signals, which are averaged by an averager 114 to provide average powers of the pseudo-residual signals. The average powers of the pseudo-residual signals are supplied to an inter-sub-band bit allocation circuit l31 in a quantization bit number adaptation circuit 115. In the inter-sub-band bit allocation circuit 131 the quantization bit number allocation to the :Erequency bands is effected based on the rate of power among -the three pseudo-residual signals and then, in a temporal bit alloca-tion circuit 132, a quantization bit allocation adaptively takes place in accordance with the temporal localization of the powers of -the pseudo-residual signals. As a time interval for changing the bit aLlocation with -time, the pitch of the input speech signal is detected and the pitch period is divided at equal intervals into sub-intervals, whi.ch are set with respect to an analysis frame. The se-tting of the sub-intervals is carried out by a sub-interval se-tting circuit 133. For each sub~interval the average power of the pseudo-residual signal is obtained and the quantization bit allocation is performed.
The average powers of the pseudo-residual signals are subjected to square root computation by a square root computation ci.rcuit 134, and the thus computed square roots and the quantization bi-t numbers de-termined by the quantization bit number allocation circuit 115 are applied to a quantiza-tion step size adaptation circui-t 116, in which the quantization step size for each band is adaptively cletermined for each sub-interval. The quantization bit numbers and -the quan-tization step sizes -thus determined are se-t in quantizers in the adaptive predictive coders 108 -to 11 a correspondiny to the sub-bands, respectively.
The quantization bi-t numbers and the quantization s-tep sizes are de-termined so tha-t quantization error power may be small. Data indica-ting tne average powers of the pseudo-residual signals from -the averager 114, the pitch period from the sub-interval setting circuit 133 and -the positions of the sub-intervals relative to the analysis frame are provided to the code combiner 121. The code combiner 121 encodes the da-ta as a whole and derives the encoded output at an output terminal 18.
Accordiny -to the prior art adaptive predictive coding system described above, a quality substantially equal to PCM (six bits) of 48 Kb/s can be ob-tained at a bit ra-te of 16 Kb/s. Since -the predictive adaptation processing and the quantization adapta-tion processing take place based on power, however, this system has the defect that the quantity of hardware needed is large. Fur-thermore, since the adaptive predictive coders 108 to 110 are the forward type, the prediction coeEficients have -to be -transmitted, resulting in -the coding efficiency being lowered by tha-tO
Incidentally, in the case of employing recursive filters in the predictors and arranging the adaptive predictive coders in a backward type, the condition for stabiliza-tion is not dependent on the prediction coefficien-ts, so that when the prediction coefficients are adapted, no stabili-ty can be ensured.
Fig. 3 illustrates an embodiment o~ the adaptive predictive coding system of the present invention. A
speech-sampled input signal in digital form from the inpu-t -terminal 1 is spli-t by band-pass filters 2, 3 and 4 into a plurality oE frequency bands. Signals of -these divided ~ 8 ~
Erequency bands are re-sampled by samplers 5, 6 and 7 lnto sub-band signals~ which are subjected -to adap-tive predictive encoding by adap-tive predic-tive coders 8, 9 and 10 to derive -therefrom residual signals.
The adaptive predic-tive coders 8, 9 and 10 are each composed of a predictor (Ps) 19 based on -the correlation be-tween adjacent sample values and a predictor (Pp) 20 based on the correlation between sample values spaced one pi-tch period apart, as shown in Fig. 4. A
prediction signal is produced by obtaining the sum of prediction values from the both predictors 19 and 20 through the use of an adder 21. A prediction residual signal is produced by obtaining the difEerence between the input signal from a terminal 22a and the prediction siynal from the adder through the use of a subtractor 22. The prediction residual signal thus obtained is quantized by a quantiæer 23 and applied via a terminal 23a to a code combiner 17 shown in Fig. 3, and the encoded output in the code combiner 17 is provided via a terminal 18 on a transmission line.
The output from the quantizer 23 is provided to an inverse quantizer 24, too, as depicted in Fig, 4, wherein it is decoded. The decoded output from the inverse quantizer 24 is added by an adder 25 to the prediction value from the predictor 19, and the added output from the adder 25 is fecl back as a local decoded signal to the predictor 19. E'urther, the output from the adder 25 is added by an adder 26 to the prediction value from the predictor 20 and the added output is fed back to the predicto~ 20.
The predictor 19 is shown to be made up of P
stages of lattice digital filters, each stage being formed by one unit time lag element 27, two multipliers 28 and two adders in a manner to compu-te the following equation.
In-ternal signals of an ith (i = 1, 2, ... P) stage fil-ter, efi~t) and ebi(-t), are ob-tained by the following equa-tion from internal signals of an (i ~ 1)th stage filter, efi 1(t) and bi-1( ) (t) = efi_1(t) + Ki(t)ebi-1( ) e (-t) = Ki(t)efi_1(t ~ ebi~1( i 1, 2, . P, eb0(t) f0( where Ki(t) (i = 1, 2, ... P) is a PARCOR coei-f:icientt which is the multiplica-tor of the two multipliers 28r and ef0(t) is the input signal to the predictor 19. A prediction signal which is provided from an adder 30 is given by the following equdtion:
xS(t) = - ~1Ki(t)ebi-1( ) On the other hand, the predictor 20 is composed oE a singLe stage of lattice digital filter, in which a time lag elemen-t 34 represents a time lag dependent on the pitch period and the sampling period in each band and is implemented by an interpolating digital filter since the time lag usually assumes a value which is a non~integral multiple of the samp]ing interval of the input signal. The prediction siynal of the predictor 20, which is provided from an adder 32 is given by the following equa-tion:
xp(t) = - h(t)x(t - d) where h(t) is a PARCOR coefficient, x(t) is a local decoded signal which is -the input to the predic-tor 20 and d is the time lag o-f the time lag element 34. The PARCOR
coefficierl-ts lii(-t) and h(t) are obtained. b~ a PARCOR
analyzer 35 frorn the in-ternal signals efi(t) and ebi(t) for each unit -timeO These predictors 19 and 20 are well-kno~n in the art.
The PARCOR coefficients are given by the following equations:
Ki(t+1 ) = tanh~.en I fi(t) - ebi(t) ¦efi(t) + ebi(t) ___ _ ~ ¦x(-t) - x(t-d)¦
h(-t+1) = tanh ~Qn ¦x(t) ~ .x(t-d)¦
where (¦ ¦) deno-tes a time average. The above equa-tions can be realized by the PARCOR analyzer 35 as shown in Fig, 5, for instance, The signals efi(t) and ebi(t) are applied to adders 36 and 37, in which efi(t) - ebi(t) and efi(t) ebi(t) are ob-tained~ and these outputs are provided to absolute value computation circuits 38 and 39, wherein their absolute values are obtainedO The absolute values are smoothed by smoothing low pass filters 40 and 41, and logarithmlc values of the smoothed ou-tputs are obtained by logarithmic value computation circuits 42 and 43. The logari-thmic values -thus obtained are applied to an adder 44 to detect the difference therebetween. The difference output thus detected is provided to a hyperbolic tanyent (tanh x) compu-tation circui-t 45 to ob-tain a hyperbolic tangent, providing the PARCOR coefficient. The abovesaid absolute value computa-tion circuits 38 and 39 may also be replaced with squarers, The logarithmic value compu-tation circuits 42 and 43 and the hyperbolic tangent computation circui-t 45 can each be so arranged as -to read out clata storecl in a read only memory.
The quan-tization bit number and -the quantization step size of the quan-tizer 23 in each of the adaptive predictive coders 8, 9 and 10 in Fig. 2 are adapted in accordance with the mean amplitude of the pseudo-residual signal in -the present invention. The pseudo-residual siynal corresponds to the residual signal provided from the subtractor 22 (Fig. 4) in the adaptive predictive coder, and the mean ampli-tude oE the pseudo residual signal has to be compu-ted prior to the activation of the adaptive predictive coder. ~s shown in Fig, 3, from the sub-band signal of the lowest band, for example, the output from the sampler 5 is detected by a pitch period de-tector 11 the pi-tch period Tp of the signal, The detection of the pitch period can be effected in a known manner. That is, an average of the absolute value of the difference between sub-band sample values in the analysis frame is obtalned by ~_~
p(~ (t) - x(t+~)¦
M - T t=0 where M is the number of samples in the frame, and the pitch period is obtained as such a value of ~ that minimizes the average p~l) within the range cf ~min ~ T < ~maX~ By the pltch period Tp thus detec-ted, the delay time of the delay element 34 in Fig. 4 is controlled, In an inverse filter circuit 12 r f~m the sub-band signals of the respective bands are ob-tained theLr pseudo-prediction residual signals, The inverse filter 12 is one tha-t corresponds to the filter used for the predictors in the adaptive predictive coders 8, 9 and 10, and it is made up of, for instance, a cascade connection - 12 ~-~ .'3~
of an inverse filter 46 based on a pi.tch predic-tion and an inverse fi.l-ter ~7 based on an adjacent correlation as shown in Fig. 6; namely, i-t is formed by -the same lattice digital filters as those employed for -the adaptive predictive coder depicted in Fig, 4. Time delay elements 48 and 52 and PARCOR analyzers 51 are identical with the time delay elements 34 and 27 and the PARCOR analyzers 35 used i,n the adaptive predic-tive coder shown in Fig. 4.
Nex-t, that one of the outputs from the inverse fil,ter 12 which corresponds -to t:he sub-band signal of t,he lowest frequency band is provided to a sub-in-terval set.ting circuit 13, wherein the position of the sub-interval is detected based on the temporal localiza-tion of -the amplitude of its pseuclo-residual signal. As shown in Fig. 7, the pi-tch period Tp is divided at equal intervals into a plurality of sub-intervals (four sub-intervals i.n Fig. 7) so that they may repeat periodically. The positions of the sub-intervals are defined by the time l.ength Td from the beginning of the analysis frame -to the beginning 11 of a first sub-interval j = 11, 12 and 13. The time length Td is set so -that the mean amplitude in the first sub-interval may become maximum. Tha-t is to say, the mean amplitude is obtained by u(Td) ~ ~ ~T Ir1( )I
where T is the first sub-interval (11, 12, 13) when the time interval is set to a cer-tain value, M is the number of samples in the first sub-interval and r1(t) i.s the residual signal, and then Td which maximizes u(Td) is obta,ined in the range of 0 ~ Td < Tp, In this way, the sub-intervals are set.
The arithmetic processing for setting the temporal ~6~ 3 position I`d can be carried ou-t following a flowchart shown in Fig. 1l, for instance. In Fig~ 11, in step S1 the mean ampli-tucle u* of -the pseudo~residual. signal of the sub-interval i.s set to _~ and ~ is se-t as a variahle of Td to 0 and, in step S2, the mean amplitude of -the pseudo-residual signal of t~e sub-interval is set to 0 and the number of sample points, M is set to 0 and, further, a saMple point t is set to 0. In step S3 i-t is checked whether the sample point t lies within the sub-intervals in which the mean amplitude is to obtain, That is, T(t) represents the following equation in which 1 is added -to an in-tegral value obtained by dividing by the length of one sub-interval, TQ = Tp/]., the remainder of a division of t by the pitch period Tp:
~'(t) = [~ P-~] ~ 1 where [ ] is a Gaussian symbol and indicates -the largest integral value in [ ]. In the case of YES in step S3, the pseudo-residual. signal ¦r1(t)¦ of that sub-interval is added to u and the number of sample points, M, is incremented by 1 in step S~, which is ollowed by step S5. In the case of NO in step S3, the operation proceeds to step S5, in which the sample point t is incremented by 1, followed by checking in step S6 whe-ther the sample point t has become a final sample point a-t the end of the analysis frame ]ength Tf. In the case of NO in step S6, the operation goes back to step S3, whereas, in -the case of YES in step S6, the operation proceeds to step S7 in which the average amplitude of the pseudo-residual signal in the same sub-interval within the analysis frame is obtained. If it is decided in the next step S8 that the average amplitude u is larger than u*, -the average ampli-tude u is set as u* and ~ is set as Td in st.ep S9, and the operatlon proceeds to step S10.
Also in -the case of NO in -the step S8, the operation proceeds to step S10, In s-tep S10 the ~ is incremen-ted by 1, which is foLlowed by step S11, in which it is checked whether the ~ has become the pitch period Tp, If so, the operation comes -to an end and the Td :Ln step Sg at that time becomes a value representing the posi-tion oE ~he sub-interval rela-tive to the analysis Erame. If not, the operation returns to step S2 and the above-described operation takes place again to obtain the -time length Td which ma.xlmizes the average amplitude of the pseudo-residual signal of -the same sub-interval, In an average amplitude computation circuit 14 shown in Fig. 3, an average amplitude of the pseudo-residual signal in each sub-interval for each band is obta:ined by the following equation:
ij Ml~ t~Tij j - 1, 2, L
where Tij and Mij indicate a jth sub-interval in an ith band and the number of samples contained in the sub-interval, respectively, N is the number of bands into which the input signal is split (N = 3 in Fig. 3) and L is the number of periods into which -the pitch period is split (L
= 4 in Fig. 7).
The arithmetic processing -for obtaining the average amplitude of the pseudo-residual signal for each sub-interval is performed, for example as shown in Fig.
12. In step S12 operation starts with initialization of i = 1 and, in s-tep S13~ uij = 0 and Mij = 0 are initiali~ed for each j = 1, 2, ... L~ followed by setting of the sample poin-t t to 0 in step S14. In step S15 -the sub~interval ~ 15 -?a j is obtained by the equa-tion of-. T(t) from T(-t - Td) and, for the sub-interval. j, u j -~ ¦ri(t1)¦ is compu-ted, that is, -the pseudo-residual signals are accumulated and, furtner, Mij is incremented by 1. In the nex-t step S16 t is incremen-ted by 1 and in step S17 it ls checked whether t has become the final sample point Tfi. If so, the opera-tion proceeds -to step S18 and if not, -the operation returns -to s-tep S15. In step S18 the accumulated uij is divided by the number of sample poin-ts in the same sub-interval, Mij, -to obtain the average amp]itude Eor each j = 1,2, ... ~" In step Slg i i.s incremented by 1 and in step S20 it i.s checked whether i is larger -than N If not, the opera-tion returns to step S12 and if so, the operation comes to an end.
A quantization bit allocation circui-t 15 in Fig.
3 performs a computation for determining the quantization bit number from the average amplitude uij The quantization bit number is determined so that a waveform distortion of a decoded signal resulting from quantization may be minimum relative to a given average bit rate The waveform distortion in the case of the quantization bit number in the band i and the sub-interval j being represented by R
is given by the following equation:
i=1 j=1 ~ ~
where c. is the ratio of -the ti.me leng-th of each sub-interval to the analysis frame length, a is a constant representing -the relationship be-tween an absolute value mean and an effective value, and I~ is a constant. The average bit rate (bits/sample) in this case is given by - l6 --3~
-the following equa-tion:
N L
i-l j -1 i j i j where wi is the ratio of the band width of each band -to the band width before spli-tting. The quan-tization bit number Rij which minimizes the waveform ~is-tortion D when the average bit rate R is made constant is givell by the following equation:
15R = maX{R + 2lg2 where uij = uij/wi and Rcij indicates the lower limit of the quantization bit number, which is selected so that -the quantization bit number in all sub-in-tervals in a certain band may not be zero. For instance, if ~ L
{Rcij} = {1, 0, 0, ..., 0} (i = 1, 2, ~0, N), one or more bits can always be allocated to the first sub~interval for all the bands.
The computation of Rij is conducted by the processing shown in Figs, 13 and 14, for instance. In step u ,.
S21 in Fig. 13 a computation u'ij = -locf2 (- ll ) is conducted for i = 1 t 2, ,,, N and j = 1, 2, O~ L and, in N L
step S22, a computation ~ wicju'ij is carried ou-t -to perform the compu-tatlon oE -the denominator in the equa-tion o~ Rij. In step S23 an operation R - d + u'lj is conducted Eor i = 1, 2, ... N and j = 1, 2, ... L. Since Rij thus computed is not always an integer, it is set to IRij and a larger one of IRij and Rcij is selected and, further, the quantization blt number R obtained by the computation is made as close to a preset average bit rate R as possible.
To this end, processing such, for example as shown in Fig.
14 is perEormed. The operation star-ts with the ini-tiali-zation of n = 0, sO = 0, Rth = 0 5 and ~ = 0.5 in s-tep .S24, and R = 0, i = 1 and j = 1 are set in steps S25, S26 and S27, respectively. Next, in step S28 Rth is added -to Rjj obtained by the processing o~ Fig. 13 and a rnaximum value of integers in Rij + Rth is set to IRij. In step S29 it is checked whe-ther IRij is smaller than Rcij and, if so, Rcij is selected as IRij in step S30 and the opera-tion proceeds to step S31 When IRij is larger than Rcij, the IRij is adopted as it is and the operation proceeds to s-tep S31~ in which the average bit rate R by the quan-tization bit number allocated by this IRij is obtained. In step S32 i is incremented by 1 and in step S33 it is checked whether j is larger than L. If not, the operation returns to step S28 and, if so~ the opera-tion proceeds to step S34, in which i is incremented by 1 In step S35 it is checked whether i is larger than N and if not, the operation proceeds to step S27 and, if so, the operation proceeds to step S36. In step S36 it is checked whether R is smaller than the preset average bit rate R and if so, it means that Rth can be increased, so that sn is set to 1 in step S37.
If not, Rth mus-t be decreased and in step S38 s is set to 1. In step S39 it is checked whether sn does not coincide with the previous one and if they are not coincident with each other, i-t is decided that sn takes +1 an~ -1 alternately And in s-tep S40 A ls set -to ~/2 and -the operation proceeds to step S~1 In the case where coincidence is detected, the o~era-tion proceeds directly to step S41 In step S41 ~ x srl is added to Rth. In step S42 it is checked whether n is larger than a prede-termined number nma . If not, the opera-tion returns to step S25 and, if so, the operation proceeds to step S~3, in which R and R are compared again to make sure that the former is smaller -than the latter. If so, the operation comes to an end and, if no-t, the opera-tion re-turns to step S25 to make R smaller than R at all times.
In a quantization step size adap-tation circuit 16 the quantization step size of a linear quantlzer is determined by the average amplitude of the residual signal and the quantization bit number.
~ij a ij Q( ij where Q(Rij) represents the quan-tization step size which minimizes a quan-tization error in the case where a signal of a zero average and unit variance is quantized by R
bits and which is determined in dependence on the probability distribution of the signal In the computation of ~ij' aQ(Rij) is stored in the forrn of a table and, by referring to this table using Rij and its output is multiplied by a and uij Q(Rij) is described, or instance, in IR~ Trans. Information Theory, Vol IT-6, pp 7-12, 1960, March. In the case of speech, a is close to the Gaussian distribution and can be set to a = ~ . It is sufficient to store aQ(Rij) in the form of a table.
In a code combiner 17, a quantized residual signal of each band and parameter information, that is, the period 13~
Tp of the sub--interval, the position Td and the average amplitude of the pseudo~residual siynal., are encoded and delivered via the termina] 18 on-to the transmisslon line.
Fig. 8 illustra-tes an embodiment for decoding a signal from an adaptive-predlctive~encoded code series, A code series applied via a terminal 141 from the transmisslon llne is separated by a code separator 55 lnto a code series of the residual slgnal and a code series of the parameter information, and the parameter information is decoded in a parameter decoder 56. In a quan-tlzatlon bit allocation circuit 57, the quan-tization bi-t Rij ls computed Erom -the decoded average amplitude uij in the manner described previous]y and, based on this, the code series of the residual signal of each band is separated into codes for each sample value in a code separator 59.
In a quantlzation step size allocation circuit 58, the quantization step size is computed by the aforementioned method from the average arnplltude uij, on the basis of whlch the residual signals are decoded by decoders 60, 61 and 62. Prediction filter circuits 63, 64 and 65 receive the residual signals and output the sub-band signals of -the respective bands, The prediction filter circuits 63, 64 and 65 are each composed of prediction filters 74 and 75 which are based on the adjacent correlation and the pitch correlation, respectivel.y~ and which are formed by a 1.attice digital filter comprising a unit time delay el.ement 81, multipliers 82 and adders 83, and a lattice digital filter comprising a time delay elemen-t 84, a multiplier 85 and an add.er 86, respectively, as is -the case with the predictors 19 and 20 in the adaptive predictive coder~ Sirnultaneously with signal filtering processing, filter coefficients Ki and h are successively estimated by PARCOR analyzers 87.
- 20 ~
The sub-band signals are sampled by interpola-tors 66, 67 and 68 with -the same sampling period as that of the input speech slynal, That is~ a sample of a zero value is inserted between sample values of each sub-band signal, Band-pass filters 69, 70 and 71 f:Llter the interpolator outputs with filters having the same characteristics as the bancl-pass Eilters 2, 3 and 4 shown in Fig, 3. I'he filter ou-tputs are added by an adder 72 and i-ts decoded signal is provided at an output terminal 73. The operations described above w:ith recjard to l~igs. 3 to 8 can also be carried out ~y the use of a microcomputer.
In Fig, 4, one of the predictors 19 and 20 may also be omi-tted~ in which case the corresponding parts in Figs. 6 and 9 are left out. Al.so it is possible to replace the preaictors 19 and 20 with each other in Fig. 4 and, in this case, the inverse filters ~6 and ~7 are replaced with each other in Fig. 6.
F~lrtllermore, in F'ig. 4, the quantized prediction re_iclu--.l signa] at the terminal 23a is decoded, from which the, PAI-'COP coe-~fici-ent is successively estimated and -the backward type adaptive predictive coders are employed, but it i5 also possible to adopt such a circuit arrangement as shown in Fig. 10 in which the PARCOR coefficient is analyzed by a predictive coefficient adap-tation circuit 83 from the input siynal applied to an input terminal 22a and the filter coefficient of a predictor 89 composed of a lattice digital filter, thereby to obtain the prediction signal. In other words, the adaptive predictive coders 8, 9 and 10 may also be the forward type. In this case, the PARCOR coefficient obtained wi.th the predictive coefficient adapta-tion circuit 88 is -transmitted to the receiving side.
The present invention is applicable, for instance, - 2l --to -the adaptive bi-t allocation -to a plurality of chanllels of a stereo signal. For example as shown in Fig, 15, signals of El channels are provided Erom input terminals 11 to 1H to sub-band split-ting circult 141, wherein the input signal of each channel is split into M frequency bands and, at the same time, they are translated in-to sub-band signals as descrihed reviouslv in conjunction ~i.th the sub-band split-.ing circult 101 'n Fig, 3. The sub-band signals of -the respective channels are subjected to adap-ti.ve predictive coding by the same circul-t arrangements as the adaptive predicti.ve coders 8, 9 and 10 constitutecl by the lattice filters as shown in Fig. 3. And residual data 143 are provicled. On the o-ther hand, the sub-band signals are provided Erom -the sub-band splitting circuit 141 to a quantiza-tion bit and step size adapta-tion circuit 144, tooc Irl th" quantizatiorl bit. ancl step size adaptation circuit ~ , the av_rage amplitude of the pseudo-residual signal for each sub--bana signal i.s obtained for each sub-in-terval, for instance, by -the same rnethod as that described previousli7 with respect to Fig. 3. Letting the average arnplitude of -the pseudo-residual signal in a jth sub-interval of 211 ith band of an hth chanr-el be represented by uhij, the quantization bit number Rhij and the quantization step size ~hij in the jth sub--interval of -the ith band of the h-th channel are obtained by the following equations:
Rhij = max {R~2log2 ~ , Rchij}
n IrI f7 ( umk ) WkC Q
m=~ 2=1 Q
where R is the average bit rate (bits/sample) fcr -the signals of -the H channe.ls.
~ 22 -u _ hi hi~ wi Qhij = a ~ Uhlj Q( hij Based on Rhij and ~hij thus obtained, the q~lantization bit number and the quantiza-tion step size in each sub-in-terval of each sub-hand signal in each channel are adaptively set.
~s has been described in the foregoing, according to the adaptive predictive coding system of the presen-t invention, the conventional band-spLitting temporal splitting methods are not employed and the signal-to-quantization noise ratio is improved abou-t 6 dB as compared with that in the adaptive predictive coding system which does not utilize -the adaptive bit allocation technique, The toll qualit~ is improved more than the abovesaid numerical value, and it has been confirmed that the toll qualit~y obtainable with the prior art can be achieved at a bi-t rate of 16 Kb/s. Furthermore, in the system of the present invention, the computation of the pitch period and the PARCOR coefficient in the inverse filter 12 and the adaptation of the quantizer are performed based on the computation for obtaining the average value of -the absolute values of the signals, so that in the case of implementing this invention system for digital signal processing, the computation word length is substantially the same as the word length of the input signal; namely, the word length can be reduced by half as compared with the word length needed in the case of effecting the adaptive quantization based on the computation of power. Moreover, the number of multiplica-tions involved is also decreased. Conversely speaking, if the word length is selected equal to that in the case of conducting the computation by computing power, then high precision coding can be effected, providing for enhanced performance. Besides, it is necessary in the prior ar-t to perEorm the bit allocation in the sub-interval by the temporal bit allocation clrcuit 132 after the compu-tation in the inter-sub-band bit allocation circui-t 131 as shown in Fig. 1. In the present invention, however, Rij can be obtained directly by the quantization bit allocation circuit 15 as described previously and, consequently, the throughput is reduced by that Further, the prior art example of Fig. 1 calls for the square root computa-tion circuit 134 for the quantiæation step size adaptation, but the present inven-tion does not require such a circuit. In the case where the computation of the PARCOR
coefficients in the adaptive predictive coder is based on the average absolute value of the signal as described in respect of the foregoing embodiment, the word length is made uniform throughou-t the sys-tem; this allows ease in performing the processing and eliminates the necessity of the multiplications and divisions which take place in the predictive coefficient computation circuit 128 in Fig 4 r reducing the scale of the hardware used as shown in Fig 6, for instance. In addition, in the case where the predictive coding and the decoding circuit are constituted by the same type of circuits as in the embodiments of Figs.
3 and 8 and the PARCOR coefficients are successively estimated from the decoded signal on either of the transmitting and the receiving side, there is no need of transmitting the PARCOR coefficients and, in consequence, the quantity of information therefor can be allocated to the quantization of the residual signal, improving the performance for coding Incidentally, since the predictor utilizing the lattice Eilter is s-table when the PARCOR
coefficients are smaller -than unity, it is sufficient at P~
the time of de-termininy the adap-tive quan-tization bit number and -the quan-tiza-tion s-tep size -to seek the condition Eor stability; thls ensures the stabili-ty of the filter in the predictive code~.
It will be apparent that many modifications and variations may be e-ffected without departing from the scope of the novel concepts of the present invention,
is given by the following equation:
i=1 j=1 ~ ~
where c. is the ratio of -the ti.me leng-th of each sub-interval to the analysis frame length, a is a constant representing -the relationship be-tween an absolute value mean and an effective value, and I~ is a constant. The average bit rate (bits/sample) in this case is given by - l6 --3~
-the following equa-tion:
N L
i-l j -1 i j i j where wi is the ratio of the band width of each band -to the band width before spli-tting. The quan-tization bit number Rij which minimizes the waveform ~is-tortion D when the average bit rate R is made constant is givell by the following equation:
15R = maX{R + 2lg2 where uij = uij/wi and Rcij indicates the lower limit of the quantization bit number, which is selected so that -the quantization bit number in all sub-in-tervals in a certain band may not be zero. For instance, if ~ L
{Rcij} = {1, 0, 0, ..., 0} (i = 1, 2, ~0, N), one or more bits can always be allocated to the first sub~interval for all the bands.
The computation of Rij is conducted by the processing shown in Figs, 13 and 14, for instance. In step u ,.
S21 in Fig. 13 a computation u'ij = -locf2 (- ll ) is conducted for i = 1 t 2, ,,, N and j = 1, 2, O~ L and, in N L
step S22, a computation ~ wicju'ij is carried ou-t -to perform the compu-tatlon oE -the denominator in the equa-tion o~ Rij. In step S23 an operation R - d + u'lj is conducted Eor i = 1, 2, ... N and j = 1, 2, ... L. Since Rij thus computed is not always an integer, it is set to IRij and a larger one of IRij and Rcij is selected and, further, the quantization blt number R obtained by the computation is made as close to a preset average bit rate R as possible.
To this end, processing such, for example as shown in Fig.
14 is perEormed. The operation star-ts with the ini-tiali-zation of n = 0, sO = 0, Rth = 0 5 and ~ = 0.5 in s-tep .S24, and R = 0, i = 1 and j = 1 are set in steps S25, S26 and S27, respectively. Next, in step S28 Rth is added -to Rjj obtained by the processing o~ Fig. 13 and a rnaximum value of integers in Rij + Rth is set to IRij. In step S29 it is checked whe-ther IRij is smaller than Rcij and, if so, Rcij is selected as IRij in step S30 and the opera-tion proceeds to step S31 When IRij is larger than Rcij, the IRij is adopted as it is and the operation proceeds to s-tep S31~ in which the average bit rate R by the quan-tization bit number allocated by this IRij is obtained. In step S32 i is incremented by 1 and in step S33 it is checked whether j is larger than L. If not, the operation returns to step S28 and, if so~ the opera-tion proceeds to step S34, in which i is incremented by 1 In step S35 it is checked whether i is larger than N and if not, the operation proceeds to step S27 and, if so, the operation proceeds to step S36. In step S36 it is checked whether R is smaller than the preset average bit rate R and if so, it means that Rth can be increased, so that sn is set to 1 in step S37.
If not, Rth mus-t be decreased and in step S38 s is set to 1. In step S39 it is checked whether sn does not coincide with the previous one and if they are not coincident with each other, i-t is decided that sn takes +1 an~ -1 alternately And in s-tep S40 A ls set -to ~/2 and -the operation proceeds to step S~1 In the case where coincidence is detected, the o~era-tion proceeds directly to step S41 In step S41 ~ x srl is added to Rth. In step S42 it is checked whether n is larger than a prede-termined number nma . If not, the opera-tion returns to step S25 and, if so, the operation proceeds to step S~3, in which R and R are compared again to make sure that the former is smaller -than the latter. If so, the operation comes to an end and, if no-t, the opera-tion re-turns to step S25 to make R smaller than R at all times.
In a quantization step size adap-tation circuit 16 the quantization step size of a linear quantlzer is determined by the average amplitude of the residual signal and the quantization bit number.
~ij a ij Q( ij where Q(Rij) represents the quan-tization step size which minimizes a quan-tization error in the case where a signal of a zero average and unit variance is quantized by R
bits and which is determined in dependence on the probability distribution of the signal In the computation of ~ij' aQ(Rij) is stored in the forrn of a table and, by referring to this table using Rij and its output is multiplied by a and uij Q(Rij) is described, or instance, in IR~ Trans. Information Theory, Vol IT-6, pp 7-12, 1960, March. In the case of speech, a is close to the Gaussian distribution and can be set to a = ~ . It is sufficient to store aQ(Rij) in the form of a table.
In a code combiner 17, a quantized residual signal of each band and parameter information, that is, the period 13~
Tp of the sub--interval, the position Td and the average amplitude of the pseudo~residual siynal., are encoded and delivered via the termina] 18 on-to the transmisslon line.
Fig. 8 illustra-tes an embodiment for decoding a signal from an adaptive-predlctive~encoded code series, A code series applied via a terminal 141 from the transmisslon llne is separated by a code separator 55 lnto a code series of the residual slgnal and a code series of the parameter information, and the parameter information is decoded in a parameter decoder 56. In a quan-tlzatlon bit allocation circuit 57, the quan-tization bi-t Rij ls computed Erom -the decoded average amplitude uij in the manner described previous]y and, based on this, the code series of the residual signal of each band is separated into codes for each sample value in a code separator 59.
In a quantlzation step size allocation circuit 58, the quantization step size is computed by the aforementioned method from the average arnplltude uij, on the basis of whlch the residual signals are decoded by decoders 60, 61 and 62. Prediction filter circuits 63, 64 and 65 receive the residual signals and output the sub-band signals of -the respective bands, The prediction filter circuits 63, 64 and 65 are each composed of prediction filters 74 and 75 which are based on the adjacent correlation and the pitch correlation, respectivel.y~ and which are formed by a 1.attice digital filter comprising a unit time delay el.ement 81, multipliers 82 and adders 83, and a lattice digital filter comprising a time delay elemen-t 84, a multiplier 85 and an add.er 86, respectively, as is -the case with the predictors 19 and 20 in the adaptive predictive coder~ Sirnultaneously with signal filtering processing, filter coefficients Ki and h are successively estimated by PARCOR analyzers 87.
- 20 ~
The sub-band signals are sampled by interpola-tors 66, 67 and 68 with -the same sampling period as that of the input speech slynal, That is~ a sample of a zero value is inserted between sample values of each sub-band signal, Band-pass filters 69, 70 and 71 f:Llter the interpolator outputs with filters having the same characteristics as the bancl-pass Eilters 2, 3 and 4 shown in Fig, 3. I'he filter ou-tputs are added by an adder 72 and i-ts decoded signal is provided at an output terminal 73. The operations described above w:ith recjard to l~igs. 3 to 8 can also be carried out ~y the use of a microcomputer.
In Fig, 4, one of the predictors 19 and 20 may also be omi-tted~ in which case the corresponding parts in Figs. 6 and 9 are left out. Al.so it is possible to replace the preaictors 19 and 20 with each other in Fig. 4 and, in this case, the inverse filters ~6 and ~7 are replaced with each other in Fig. 6.
F~lrtllermore, in F'ig. 4, the quantized prediction re_iclu--.l signa] at the terminal 23a is decoded, from which the, PAI-'COP coe-~fici-ent is successively estimated and -the backward type adaptive predictive coders are employed, but it i5 also possible to adopt such a circuit arrangement as shown in Fig. 10 in which the PARCOR coefficient is analyzed by a predictive coefficient adap-tation circuit 83 from the input siynal applied to an input terminal 22a and the filter coefficient of a predictor 89 composed of a lattice digital filter, thereby to obtain the prediction signal. In other words, the adaptive predictive coders 8, 9 and 10 may also be the forward type. In this case, the PARCOR coefficient obtained wi.th the predictive coefficient adapta-tion circuit 88 is -transmitted to the receiving side.
The present invention is applicable, for instance, - 2l --to -the adaptive bi-t allocation -to a plurality of chanllels of a stereo signal. For example as shown in Fig, 15, signals of El channels are provided Erom input terminals 11 to 1H to sub-band split-ting circult 141, wherein the input signal of each channel is split into M frequency bands and, at the same time, they are translated in-to sub-band signals as descrihed reviouslv in conjunction ~i.th the sub-band split-.ing circult 101 'n Fig, 3. The sub-band signals of -the respective channels are subjected to adap-ti.ve predictive coding by the same circul-t arrangements as the adaptive predicti.ve coders 8, 9 and 10 constitutecl by the lattice filters as shown in Fig. 3. And residual data 143 are provicled. On the o-ther hand, the sub-band signals are provided Erom -the sub-band splitting circuit 141 to a quantiza-tion bit and step size adapta-tion circuit 144, tooc Irl th" quantizatiorl bit. ancl step size adaptation circuit ~ , the av_rage amplitude of the pseudo-residual signal for each sub--bana signal i.s obtained for each sub-in-terval, for instance, by -the same rnethod as that described previousli7 with respect to Fig. 3. Letting the average arnplitude of -the pseudo-residual signal in a jth sub-interval of 211 ith band of an hth chanr-el be represented by uhij, the quantization bit number Rhij and the quantization step size ~hij in the jth sub--interval of -the ith band of the h-th channel are obtained by the following equations:
Rhij = max {R~2log2 ~ , Rchij}
n IrI f7 ( umk ) WkC Q
m=~ 2=1 Q
where R is the average bit rate (bits/sample) fcr -the signals of -the H channe.ls.
~ 22 -u _ hi hi~ wi Qhij = a ~ Uhlj Q( hij Based on Rhij and ~hij thus obtained, the q~lantization bit number and the quantiza-tion step size in each sub-in-terval of each sub-hand signal in each channel are adaptively set.
~s has been described in the foregoing, according to the adaptive predictive coding system of the presen-t invention, the conventional band-spLitting temporal splitting methods are not employed and the signal-to-quantization noise ratio is improved abou-t 6 dB as compared with that in the adaptive predictive coding system which does not utilize -the adaptive bit allocation technique, The toll qualit~ is improved more than the abovesaid numerical value, and it has been confirmed that the toll qualit~y obtainable with the prior art can be achieved at a bi-t rate of 16 Kb/s. Furthermore, in the system of the present invention, the computation of the pitch period and the PARCOR coefficient in the inverse filter 12 and the adaptation of the quantizer are performed based on the computation for obtaining the average value of -the absolute values of the signals, so that in the case of implementing this invention system for digital signal processing, the computation word length is substantially the same as the word length of the input signal; namely, the word length can be reduced by half as compared with the word length needed in the case of effecting the adaptive quantization based on the computation of power. Moreover, the number of multiplica-tions involved is also decreased. Conversely speaking, if the word length is selected equal to that in the case of conducting the computation by computing power, then high precision coding can be effected, providing for enhanced performance. Besides, it is necessary in the prior ar-t to perEorm the bit allocation in the sub-interval by the temporal bit allocation clrcuit 132 after the compu-tation in the inter-sub-band bit allocation circui-t 131 as shown in Fig. 1. In the present invention, however, Rij can be obtained directly by the quantization bit allocation circuit 15 as described previously and, consequently, the throughput is reduced by that Further, the prior art example of Fig. 1 calls for the square root computa-tion circuit 134 for the quantiæation step size adaptation, but the present inven-tion does not require such a circuit. In the case where the computation of the PARCOR
coefficients in the adaptive predictive coder is based on the average absolute value of the signal as described in respect of the foregoing embodiment, the word length is made uniform throughou-t the sys-tem; this allows ease in performing the processing and eliminates the necessity of the multiplications and divisions which take place in the predictive coefficient computation circuit 128 in Fig 4 r reducing the scale of the hardware used as shown in Fig 6, for instance. In addition, in the case where the predictive coding and the decoding circuit are constituted by the same type of circuits as in the embodiments of Figs.
3 and 8 and the PARCOR coefficients are successively estimated from the decoded signal on either of the transmitting and the receiving side, there is no need of transmitting the PARCOR coefficients and, in consequence, the quantity of information therefor can be allocated to the quantization of the residual signal, improving the performance for coding Incidentally, since the predictor utilizing the lattice Eilter is s-table when the PARCOR
coefficients are smaller -than unity, it is sufficient at P~
the time of de-termininy the adap-tive quan-tization bit number and -the quan-tiza-tion s-tep size -to seek the condition Eor stability; thls ensures the stabili-ty of the filter in the predictive code~.
It will be apparent that many modifications and variations may be e-ffected without departing from the scope of the novel concepts of the present invention,
Claims (15)
1. An adaptive predictive processing system comprising:
splitting means for splitting an input signal into a plurality-of frequency bands;
sub-band translating means for translating into a sub-band signal the signal of each of the frequency bands split by the splitting means;
adaptive predictive coding means for subjecting each sub-band signal to adaptive predictive coding through utilization of a lattice filter in a predictor;
pseudo-residual signal generating means for generating, for each sub-band signal, a pseudo-residual signal corresponding to a residual signal in the adaptive predictive coding means;
sub-interval setting means for dividing the pitch period of the input signal at equal intervals into a plurality of sub-intervals and setting the sub-intervals with respect to an analysis frame;
amplitude detecting means for detecting an average amplitude of the pseudo-residual signal of each sub-band signal for each sub-interval; and quantization step size and quantization bit number allocation means for adaptively computing a quantization step size and a quantization bit number for a prediction residual signal in the adaptive predictive coding means from the average amplitude of the corresponding pseudo-residual signal for each sub-interval.
splitting means for splitting an input signal into a plurality-of frequency bands;
sub-band translating means for translating into a sub-band signal the signal of each of the frequency bands split by the splitting means;
adaptive predictive coding means for subjecting each sub-band signal to adaptive predictive coding through utilization of a lattice filter in a predictor;
pseudo-residual signal generating means for generating, for each sub-band signal, a pseudo-residual signal corresponding to a residual signal in the adaptive predictive coding means;
sub-interval setting means for dividing the pitch period of the input signal at equal intervals into a plurality of sub-intervals and setting the sub-intervals with respect to an analysis frame;
amplitude detecting means for detecting an average amplitude of the pseudo-residual signal of each sub-band signal for each sub-interval; and quantization step size and quantization bit number allocation means for adaptively computing a quantization step size and a quantization bit number for a prediction residual signal in the adaptive predictive coding means from the average amplitude of the corresponding pseudo-residual signal for each sub-interval.
2. An adaptive predictive processing system according to claim 1 wherein the quantization step size and quantization bit number allocation means is means for selecting, as the quantization bit number Rij in a jth sub-interval of an ith one of the plurality of sub-band signals, a layer one of where R (bits/sample) is the average quantization bit number, , uij is the average amplitude of the pseudo-residual signal of the corresponding sub-band signal and wi is a band width ratio (the ratio of the sub band signal band to the entire band being 1).
3. on adaptive predictive processing system according to claim 2 wherein the quantization step size and quantization bit number allocation means determines the quantization step size .DELTA.ij in the jth sub-interval of the ith sub-band signal by auij x Q(Rij), where a is a constant representing the relation between an average absolute value and an effective value and Q(Rij) is a quantization step size which minimizes a quantization error in the case of quantizing a signal of an average 0 and variance 1 by Rij bits and which is determined depending on the probability distribution of the signal.
4. An adaptive predictive processing system according to any one of claims 1 to 3 wherein the adaptive predictive coding means is backward type adaptive predictive coding means which locally decodes the adaptive predictive coded output, subjects the locally decoded signal to lattice type PARCOR analysis and adaptively controls the constant of the filter of the predictor using the analysis result as a prediction coefficient.
5. An adaptive predictive processing system according to any one of claims 1 to 3 wherein the adaptive predictive coding means is forward type adaptive pre-dictive coding means which subjects the input signal to lattice type PARCOR analysis and adaptively controls the constant of the filter of the predictor using the analysis result as a prediction coefficient.
6. An adaptive predictive processing system according to claim 1, 2 or 3, wherein the adaptive pre-dictive coding means is backward type adaptive predictive coding means which locally decodes the adaptive predictive coded output, subjects the locally decoded signal to lattice type PARCOR analysis and adaptively controls the constant of the filter of the predictor using the analysis result as a prediction coefficient wherein the predictor generates a prediction signal based on the correlation between adjacent sample values of the input signal.
7. An adaptive predictive processing system according to claim 1, 2 or 3, wherein the adaptive predictive coding means is backward type adaptive predictive coding means which locally decodes the adaptive predictive coded output, subjects the locally decoded signal to lattice type PARCOR analysis and adaptively controls the constant of the filter of the predictor using the analysis result as a prediction coefficient and wherein the predictor generates a prediction signal based on the correlation between adjacent sample values of the input signal and wherein the predictor generates the prediction signal based on the correlation between sample values spaced apart the pitch period of the input signal in addition to the correlation between the adjacent sample values of the input signal.
3. An adaptive predictive processing system according to claim 1, 2 or 3, wherein the adaptive predictive coding means is backward type adaptive predictive coding means which locally decodes the adaptive predictive coded output, subjects the locally decoded signal to lattice type PARCOR analysis and adaptively controls the constant of the filter of the predictor using the analysis result as a prediction coefficient and wherein the predictor generates a prediction signal based on the correlation between sample values spaced apart the pitch period of the input signal.
9. An adaptive predictive processing system according to claim 1, 2 or 3 wherein the adaptive pre-dictive coding means is backward type adaptive predictive coding means which locally decodes the adaptive predictive coded output, subjects the locally decoded signal to lattice type PARCOR analysis and adaptively controls the constant of the filter of the predictor using the analysis result as a prediction coefficient and wherein the pseudo-residual signal generating means is a filter of the same construction as the predictor used in the adaptive predictive coding means.
10. An adaptive predictive processing system according to claim 1, 2 or 3 wherein the adaptive pre-dictive coding means is backward type adaptive predictive coding means which locally decodes the adaptive predictive coded output, subjects the locally decoded signal to lattice type PARCOR analysis and adaptively controls the constant of the filter of the predictor using the analysis result as a prediction coefficient and wherein means for performing the lattice type PARCOR analysis comprises means for obtaining the sum of two signals and the difference therebetween at each stage of the lattice type filter, means for obtaining the absolute values of the sum and the difference signal, means for smoothing the absolute values of the sum and the difference signal, means for computing a logarithmic value of each of the smoothed values, and means for computing a hyperbolic tangent of the difference between the logarithmic values to output the PARCOR coefficient.
11. An adaptive predictive processing system according to claim 1, 2 or 3, wherein the adaptive predictive coding means is backward type adaptive pre-dictive coding means which locally decodes the adaptive predictive coded output, subjects the locally decoded signal to lattice type PARCOR analysis and adaptively controls the constant of the filter of the predictor using the analysis result as a prediction coefficient and wherein the sub-interval setting means comprises pitch detecting means for detecting the pitch period of the input signal, and means for determining the position of one of the sub-intervals in the detected pitch period with respect to the analysis frame so that the average amplitude of the sub-interval may become the largest.
12. An adaptive predictive processing system which decodes a speech signal based on a code series of a residual signal, an average amplitude signal of a pseudo-residual signal for each sub-interval of each band and sub-interval information, comprising:
quantization bit number allocating means for computing a quantization bit number of each sub-interval of each band by a predetermined method from the average amplitude signal;
code separating means for separating the code series of the residual signal for each band based on the quantization bit number of each sub-interval of each band obtained with the quantization bit number allocating means;
quantization step size allocating means for obtaining a quantization step size of each sub-interval of each band by a predetermined method from the quantization bit number of each sub-interval of each band obtained with the quantization bit number allocating means and the average amplitude signal;
decoding means for decoding the separated code series of the residual signal through using the quantization step size and the quantization bit number corresponding thereto;
predictive decoding means for the decoded residual signal. of each band into the sub band signal through using a lattice type filter in a predictor; and frequency translating means for translating the decoded sub-band signal into the corresponding frequency band,
quantization bit number allocating means for computing a quantization bit number of each sub-interval of each band by a predetermined method from the average amplitude signal;
code separating means for separating the code series of the residual signal for each band based on the quantization bit number of each sub-interval of each band obtained with the quantization bit number allocating means;
quantization step size allocating means for obtaining a quantization step size of each sub-interval of each band by a predetermined method from the quantization bit number of each sub-interval of each band obtained with the quantization bit number allocating means and the average amplitude signal;
decoding means for decoding the separated code series of the residual signal through using the quantization step size and the quantization bit number corresponding thereto;
predictive decoding means for the decoded residual signal. of each band into the sub band signal through using a lattice type filter in a predictor; and frequency translating means for translating the decoded sub-band signal into the corresponding frequency band,
13. An adaptive predictive processing system comprising:
splitting means for splitting input signals of a plurality of channels into a plurality of frequency bands;
sub-band translating means for translating the signal of each of the frequency bands split by the splitting means into a sub-band signal;
adaptive predictive coding means for subjecting the sub-band signal of each channel to adaptive predictive coding through using a lattice type filter in a predictor;
pseudo-residual signal generating means for generating a pseudo-residual signal corresponding to a residual signal in the adaptive predictive coding means for each sub-band signal of each channel;
sub-interval setting means for dividing the pitch period of each input signal of each channel at equal intervals into a plurality of sub-intervals and setting them with respect to an analysis frame;
amplitude detecting means for detecting an average amplitude of the pseudo-residual signal of each sub-band signal of each channel for each sub-interval; and quantization step size and quantization bit number adaptation means for adaptively computing, for each sub-interval, a quantization step size and a quantization bit number for a prediction residual signal in the adaptive predictive coding means from the average amplitude of the corresponding residual signal.
splitting means for splitting input signals of a plurality of channels into a plurality of frequency bands;
sub-band translating means for translating the signal of each of the frequency bands split by the splitting means into a sub-band signal;
adaptive predictive coding means for subjecting the sub-band signal of each channel to adaptive predictive coding through using a lattice type filter in a predictor;
pseudo-residual signal generating means for generating a pseudo-residual signal corresponding to a residual signal in the adaptive predictive coding means for each sub-band signal of each channel;
sub-interval setting means for dividing the pitch period of each input signal of each channel at equal intervals into a plurality of sub-intervals and setting them with respect to an analysis frame;
amplitude detecting means for detecting an average amplitude of the pseudo-residual signal of each sub-band signal of each channel for each sub-interval; and quantization step size and quantization bit number adaptation means for adaptively computing, for each sub-interval, a quantization step size and a quantization bit number for a prediction residual signal in the adaptive predictive coding means from the average amplitude of the corresponding residual signal.
14. An adaptive predictive processing system according to claim 13 wherein the quantization step size and quantization bit number adaptation means is means for selecting, as the quantization bit number Rhij in a jth sub-interval of an ith sub band signal corresponding to an hth one of the input signals of the plurality of channels, a larger one of where R (bits/sample) is an average quantization bit number of an H channel signal and , Uhij being an average amplitude of the pseudo-residual signal of the corresponding sub-band signal and wi being a band width ratio (the ratio of the sub-band signal to the entire band being 1).
15. An adaptive predictive processing system according to claim 14 wherein the quantization step size and quantization bit number adaptation means determines the quantization step size .DELTA.hij in the jth sub-interval of the ith sub-band signal of the hth channel by a ? uhij ?
Q(Rhij), where a is a constant representing the relation between an average absolute value and an effective value and Q(Rhij) is a quantization step size which minimizes a quantization error in the case of quantizing a signal of a zero average and unit variance by Rhij bits and which is determined depending on the probability distribution of the signal.
Q(Rhij), where a is a constant representing the relation between an average absolute value and an effective value and Q(Rhij) is a quantization step size which minimizes a quantization error in the case of quantizing a signal of a zero average and unit variance by Rhij bits and which is determined depending on the probability distribution of the signal.
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| JP177564/1981 | 1981-11-04 | ||
| JP56177564A JPS5921039B2 (en) | 1981-11-04 | 1981-11-04 | Adaptive predictive coding method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CA1190323A true CA1190323A (en) | 1985-07-09 |
Family
ID=16033161
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CA000410713A Expired CA1190323A (en) | 1981-11-04 | 1982-09-02 | Adaptive predictive processing system |
Country Status (3)
| Country | Link |
|---|---|
| US (1) | US4538234A (en) |
| JP (1) | JPS5921039B2 (en) |
| CA (1) | CA1190323A (en) |
Families Citing this family (57)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US4691292A (en) * | 1983-04-13 | 1987-09-01 | Rca Corporation | System for digital multiband filtering |
| CA1253255A (en) * | 1983-05-16 | 1989-04-25 | Nec Corporation | System for simultaneously coding and decoding a plurality of signals |
| US4672670A (en) * | 1983-07-26 | 1987-06-09 | Advanced Micro Devices, Inc. | Apparatus and methods for coding, decoding, analyzing and synthesizing a signal |
| US4701953A (en) * | 1984-07-24 | 1987-10-20 | The Regents Of The University Of California | Signal compression system |
| GB8421498D0 (en) * | 1984-08-24 | 1984-09-26 | British Telecomm | Frequency domain speech coding |
| US4622680A (en) * | 1984-10-17 | 1986-11-11 | General Electric Company | Hybrid subband coder/decoder method and apparatus |
| US4675863A (en) | 1985-03-20 | 1987-06-23 | International Mobile Machines Corp. | Subscriber RF telephone system for providing multiple speech and/or data signals simultaneously over either a single or a plurality of RF channels |
| US4654863A (en) * | 1985-05-23 | 1987-03-31 | At&T Bell Laboratories | Wideband adaptive prediction |
| JPH07123227B2 (en) * | 1985-08-30 | 1995-12-25 | 日本電気株式会社 | Speech coder |
| GB2182795B (en) * | 1985-11-12 | 1988-10-05 | Nat Res Dev | Apparatus and methods for speech analysis |
| US4831550A (en) * | 1986-03-27 | 1989-05-16 | International Business Machines Corporation | Apparatus and method for estimating, from sparse data, the probability that a particular one of a set of events is the next event in a string of events |
| US4899384A (en) * | 1986-08-25 | 1990-02-06 | Ibm Corporation | Table controlled dynamic bit allocation in a variable rate sub-band speech coder |
| DE3639753A1 (en) * | 1986-11-21 | 1988-06-01 | Inst Rundfunktechnik Gmbh | METHOD FOR TRANSMITTING DIGITALIZED SOUND SIGNALS |
| GB8628046D0 (en) * | 1986-11-24 | 1986-12-31 | British Telecomm | Transmission system |
| NL8700985A (en) * | 1987-04-27 | 1988-11-16 | Philips Nv | SYSTEM FOR SUB-BAND CODING OF A DIGITAL AUDIO SIGNAL. |
| US4979188A (en) * | 1988-04-29 | 1990-12-18 | Motorola, Inc. | Spectrally efficient method for communicating an information signal |
| US4956871A (en) * | 1988-09-30 | 1990-09-11 | At&T Bell Laboratories | Improving sub-band coding of speech at low bit rates by adding residual speech energy signals to sub-bands |
| US4965830A (en) * | 1989-01-17 | 1990-10-23 | Unisys Corp. | Apparatus for estimating distortion resulting from compressing digital data |
| IT1232084B (en) * | 1989-05-03 | 1992-01-23 | Cselt Centro Studi Lab Telecom | CODING SYSTEM FOR WIDE BAND AUDIO SIGNALS |
| US5434948A (en) * | 1989-06-15 | 1995-07-18 | British Telecommunications Public Limited Company | Polyphonic coding |
| US5235671A (en) * | 1990-10-15 | 1993-08-10 | Gte Laboratories Incorporated | Dynamic bit allocation subband excited transform coding method and apparatus |
| JP2833212B2 (en) * | 1990-11-29 | 1998-12-09 | 松下電器産業株式会社 | Bit allocation method for band division coding |
| US5317672A (en) * | 1991-03-05 | 1994-05-31 | Picturetel Corporation | Variable bit rate speech encoder |
| EP0551705A3 (en) * | 1992-01-15 | 1993-08-18 | Ericsson Ge Mobile Communications Inc. | Method for subbandcoding using synthetic filler signals for non transmitted subbands |
| US5384793A (en) * | 1992-01-28 | 1995-01-24 | Ericsson Ge Mobile Communications Inc. | Fading and random pattern error protection method for dynamic bit allocation sub-band coding |
| CA2088082C (en) * | 1992-02-07 | 1999-01-19 | John Hartung | Dynamic bit allocation for three-dimensional subband video coding |
| JP2976701B2 (en) * | 1992-06-24 | 1999-11-10 | 日本電気株式会社 | Quantization bit number allocation method |
| KR100287494B1 (en) * | 1993-06-30 | 2001-04-16 | 이데이 노부유끼 | Digital signal encoding method and apparatus, decoding method and apparatus and recording medium of encoded signal |
| JPH09503630A (en) * | 1993-07-07 | 1997-04-08 | ピクチャーテル コーポレイション | Constant bit rate speech encoder / decoder |
| US5664057A (en) * | 1993-07-07 | 1997-09-02 | Picturetel Corporation | Fixed bit rate speech encoder/decoder |
| US5546383A (en) | 1993-09-30 | 1996-08-13 | Cooley; David M. | Modularly clustered radiotelephone system |
| EP0692881B1 (en) * | 1993-11-09 | 2005-06-15 | Sony Corporation | Quantization apparatus, quantization method, high efficiency encoder, high efficiency encoding method, decoder, high efficiency encoder and recording media |
| JP3563756B2 (en) * | 1994-02-04 | 2004-09-08 | 富士通株式会社 | Speech synthesis system |
| DE69429620D1 (en) * | 1994-10-28 | 2002-02-14 | Rai Radiotelevisione Italiana | Subband coding with pitch based prediction coding in each individual subband |
| JP3749752B2 (en) * | 1995-03-24 | 2006-03-01 | アイティーティー・マニュファクチャリング・エンタープライジズ・インコーポレーテッド | Block adaptive differential pulse code modulation system |
| US5710781A (en) * | 1995-06-02 | 1998-01-20 | Ericsson Inc. | Enhanced fading and random pattern error protection for dynamic bit allocation sub-band coding |
| US5956674A (en) * | 1995-12-01 | 1999-09-21 | Digital Theater Systems, Inc. | Multi-channel predictive subband audio coder using psychoacoustic adaptive bit allocation in frequency, time and over the multiple channels |
| JP4121578B2 (en) * | 1996-10-18 | 2008-07-23 | ソニー株式会社 | Speech analysis method, speech coding method and apparatus |
| FI114248B (en) * | 1997-03-14 | 2004-09-15 | Nokia Corp | Method and apparatus for audio coding and audio decoding |
| TW390082B (en) * | 1998-05-26 | 2000-05-11 | Koninkl Philips Electronics Nv | Transmission system with adaptive channel encoder and decoder |
| US6961432B1 (en) * | 1999-04-29 | 2005-11-01 | Agere Systems Inc. | Multidescriptive coding technique for multistream communication of signals |
| JP2001175298A (en) * | 1999-12-13 | 2001-06-29 | Fujitsu Ltd | Noise suppression device |
| KR20020006164A (en) * | 2000-07-11 | 2002-01-19 | 송문섭 | Method for extracting linear predictive coding coefficients through lattice method in vocal track encoding |
| JP3859462B2 (en) * | 2001-05-18 | 2006-12-20 | 株式会社東芝 | Prediction parameter analysis apparatus and prediction parameter analysis method |
| US7240001B2 (en) | 2001-12-14 | 2007-07-03 | Microsoft Corporation | Quality improvement techniques in an audio encoder |
| US6934677B2 (en) | 2001-12-14 | 2005-08-23 | Microsoft Corporation | Quantization matrices based on critical band pattern information for digital audio wherein quantization bands differ from critical bands |
| JP4676140B2 (en) * | 2002-09-04 | 2011-04-27 | マイクロソフト コーポレーション | Audio quantization and inverse quantization |
| US7502743B2 (en) * | 2002-09-04 | 2009-03-10 | Microsoft Corporation | Multi-channel audio encoding and decoding with multi-channel transform selection |
| US7299190B2 (en) * | 2002-09-04 | 2007-11-20 | Microsoft Corporation | Quantization and inverse quantization for audio |
| US7567900B2 (en) * | 2003-06-11 | 2009-07-28 | Panasonic Corporation | Harmonic structure based acoustic speech interval detection method and device |
| US7460990B2 (en) | 2004-01-23 | 2008-12-02 | Microsoft Corporation | Efficient coding of digital media spectral data using wide-sense perceptual similarity |
| US7885819B2 (en) | 2007-06-29 | 2011-02-08 | Microsoft Corporation | Bitstream syntax for multi-process audio decoding |
| US8249883B2 (en) | 2007-10-26 | 2012-08-21 | Microsoft Corporation | Channel extension coding for multi-channel source |
| JPWO2010140590A1 (en) * | 2009-06-03 | 2012-11-22 | 日本電信電話株式会社 | PARCOR coefficient quantization method, PARCOR coefficient quantization apparatus, program, and recording medium |
| DE102010006573B4 (en) * | 2010-02-02 | 2012-03-15 | Rohde & Schwarz Gmbh & Co. Kg | IQ data compression for broadband applications |
| JP6039678B2 (en) * | 2011-10-27 | 2016-12-07 | エルジー エレクトロニクス インコーポレイティド | Audio signal encoding method and decoding method and apparatus using the same |
| US20150025894A1 (en) * | 2013-07-16 | 2015-01-22 | Electronics And Telecommunications Research Institute | Method for encoding and decoding of multi channel audio signal, encoder and decoder |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| FR2412987A1 (en) * | 1977-12-23 | 1979-07-20 | Ibm France | PROCESS FOR COMPRESSION OF DATA RELATING TO THE VOICE SIGNAL AND DEVICE IMPLEMENTING THIS PROCEDURE |
| US4184049A (en) * | 1978-08-25 | 1980-01-15 | Bell Telephone Laboratories, Incorporated | Transform speech signal coding with pitch controlled adaptive quantizing |
| US4401855A (en) * | 1980-11-28 | 1983-08-30 | The Regents Of The University Of California | Apparatus for the linear predictive coding of human speech |
-
1981
- 1981-11-04 JP JP56177564A patent/JPS5921039B2/en not_active Expired
-
1982
- 1982-09-02 CA CA000410713A patent/CA1190323A/en not_active Expired
- 1982-09-02 US US06/414,266 patent/US4538234A/en not_active Expired - Lifetime
Also Published As
| Publication number | Publication date |
|---|---|
| JPS5921039B2 (en) | 1984-05-17 |
| US4538234A (en) | 1985-08-27 |
| JPS5878200A (en) | 1983-05-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CA1190323A (en) | Adaptive predictive processing system | |
| US5260980A (en) | Digital signal encoder | |
| US4617676A (en) | Predictive communication system filtering arrangement | |
| US4811396A (en) | Speech coding system | |
| US4956871A (en) | Improving sub-band coding of speech at low bit rates by adding residual speech energy signals to sub-bands | |
| JP3178026B2 (en) | Digital signal encoding device and decoding device | |
| CA1243121A (en) | Multi-pulse excitation linear-predictive speech coder | |
| US4680797A (en) | Secure digital speech communication | |
| US6345246B1 (en) | Apparatus and method for efficiently coding plural channels of an acoustic signal at low bit rates | |
| US5699484A (en) | Method and apparatus for applying linear prediction to critical band subbands of split-band perceptual coding systems | |
| US5089818A (en) | Method of transmitting or storing sound signals in digital form through predictive and adaptive coding and installation therefore | |
| US4685115A (en) | Apparatus for transmitting digital signals | |
| JPH0435123A (en) | Digital signal encoder | |
| US6029125A (en) | Reducing sparseness in coded speech signals | |
| US8160870B2 (en) | Method, apparatus, program, and recording medium for long-term prediction coding and long-term prediction decoding | |
| US4354057A (en) | Predictive signal coding with partitioned quantization | |
| US5311549A (en) | Method and system for processing the pre-echoes of an audio-digital signal coded by frequency transformation | |
| US4319082A (en) | Adaptive prediction differential-PCM transmission method and circuit using filtering by sub-bands and spectral analysis | |
| CA1219373A (en) | Adaptive differential pulse code modulation decoding circuit | |
| USRE32124E (en) | Predictive signal coding with partitioned quantization | |
| US4939749A (en) | Differential encoder with self-adaptive predictive filter and a decoder suitable for use in connection with such an encoder | |
| JPS5887936A (en) | Digital information transmission system | |
| CA1334688C (en) | Multi-pulse type encoder having a low transmission rate | |
| US5561609A (en) | Transmission system with reconstruction of missing signal samples | |
| CA2246901C (en) | A method for improving performance of a voice coder |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| MKEX | Expiry |