Method and device for coding an audio signal by "forward" and "backward" lpc analysis

Translation of description of equivalent WO 9847134 A1

METHOD AND DEVICE FOR ENCODING AN ANALYSIS BY audio signal LPC "FRONT" and "BACK"

The invention relates to a method and device for coding an audio signal, such as speech signal by LPC analysis "front" and "rear".

At present, the audio signal encoding techniques, including speech signals, aim to enable the transmission of these signals in digital form, in conditions of reduced transmission rate, in order, of ensure appropriate management of these signals transmission systems taking into account 1 large increase in transactions between users. Among the encoding techniques used, one designated by LPC analysis, for "Linear Predictive Coding" in Anglo-Saxon language, consists in performing a linear prediction of the audio signal to be coded, coding being time achieved by means of a prediction filter linear applied to successive blocks of the signal.

In the above techniques, that known under the name CELP coding, for "Code Exclted Linear Prediction", is the most widespread and the one of the most efficient. Other techniques, such as the technique known as MP-LPC, for "Draw ulti Li-near Predictive Coding", or VSELP technique for "Vector Sum Exclted Linear Prediction" in Anglo-Saxon language, are relatively close coding CELP.

aforementioned coding techniques are so-called "analysis by synthesis". In particular, they permit, for audio signals belonging to the phone frequency band, reducing the transmission rate of the signal 64 b / s (PCM) to 16 kb / s using the coding technique CELP, and even up to 8 Mb / s in the case of encoders employing the latest developments of this coding technique, without degradation perceived quality of speech returned after transmission and decoding.

A particularly important area of ​​application of these coding techniques is, in particular, that of mobile telephony. In this scope, the necessary limitation of the frequency band allocated to each mobile operator and the rapid increase in users subscribing necessitate a corresponding decrease in the coding rate, while the requirements of users for speech quality are increasing. Other areas of application of such coding techniques include, for example, storage of digital data representing these signals on the storage media, high quality telephony applications video- or audio-conferencing, multimedia, or digital satellite transmissions.

The linear prediction filters used in the techniques mentioned above are obtained using a so-called analysis module "LPC analysis" operating on successive blocks of the digital signal. These filters are capable, according to the scanning order, that is to say according to the number of filter coefficients to model more or less accurately the outline of the signal frequency spectrum to be encoded. In the case of a speech signal, these contours are called formants.

However, for a good coding, required by most current applications, the defined filter is not enough to model the signal perfectly. It is then necessary to carry out encoding of the linear prediction residue. Such a procedure relating to the LP residue is especially implemented by the LD-CELP coding technique for Lovr Delay CELP language Anglo-Saxon, previously mentioned in the description. The residual signal is in this case modeled by a waveform extracted from a stochastic codebook and multiplied by a gain value. The coding technique MP-LPC, for example, models this residue with variable position pulses assigned to respective gain values, while the VSELP coding technique performs this modeling by a linear combination of appropriate directories extracts pulse vectors .

A didactic recall procedure 1 'LPC analysis including the LPC analysis "backward" LPC analysis and "forward" LPC analysis or "backward" LPC analysis and "forward", respectively, in Anglo-Saxon language, will first of all given hereinafter.

The overall envelope of the frequency spectrum is odélisée through a synthesis filter in the short term, constituting the LPC filter, whose coefficients are evaluated using linear prediction of the speech signal to be encoded. This LPC filter, autoregressive filter has a transfer function of the form, relation (1)^~ :

F

A (z) • = 1 - Σ az^"1 where p denotes the number of coefficients a₁ the filter and the order of the linear prediction implementation, z designating the variable of the z transform of the frequency space.

A coefficient evaluation method₁ is to apply a criterion for minimizing the energy of the speech signal of the prediction error signal on the analysis length thereof.

The length of analysis for a digital speech signal formed of successive samples is in practice a number N of these samples, constituting a coding frame. The energy of the prediction error signal then satisfies the relationship (2):

N p

Ep = L (s (n) - The SCN-EI i))² where s (n) denotes the sample of rank n in the frame of N samples. In a block encoding process, the encoding frame may advantageously be divided into several sub frames or adjacent LPC blocks. N analysis then length exceeds the length of each block to allow the consideration of a number of past samples and, where appropriate, futures, the medium and the price of appropriate coding delays.

The analysis is called "forward" LPC when the LPC analysis process is conducted on the block of the current frame of the speech signal to be coded, the coding at the coder involved in "real time", that is to -dire during the block in the current frame only one processing delay introduced near by calculation of the filter coefficients. This analysis involves transmission of the calculated values ​​of the coefficients of the filters at the decoder.

The LPC analysis "rear" implemented in the LD-CELP encoder at 16 kb / s is subject to the ITU-T G728 standard. This analytical technique is to perform the LPC analysis, not on or the block of the current frame of the speech signal to be encoded, but on the synthesis signal. We now understand that this LPC analysis is performed actually on the synthesis signal of the block preceding the current block, because this signal is available simultaneously at the encoder and the decoder. This simultaneous operation at the encoder and the decoder and prevents the transmission of the encoder to the decoder of the value obtained at the encoder, the LPC filter coefficients. For this reason, the analysis "backward" LPC frees the transmission rate, the rate thus released can be used for example to enrich the excitation dictionaries in the case of CELP. The analysis "backward" LPC further allows an increase in the scanning order, the number of LPC filter coefficients up to 50 in the case of an LD-CELP encoder against 10 coefficients for most encoders using implements a LPC analysis "before". Thus, proper operation of the LPC analysis "back" requires the following conditions:

- Good quality of the synthesis signal, very close to the speech signal to be coded, which implies a sufficiently high coding rate greater than 13 kbit / s in view of the current quality of CELP coders;

- Frame and reduced length block because of the delay of a block between signal analyzed and signal coding. The frame length and block must be small compared to the average stationary time of the speech signal to be coded;

- Loyalty transmission and respect for the integrity of data transmitted between encoder and decoder, by introducing few transmission errors. Once the synthesis signals differ significantly from the speech signal to be encoded, encoder and decoder no longer calculate the same filter and discrepancies can occur in the absence of any chance of a return to identity sensitive filters calculated co deur or decoder.

Because of the advantages and disadvantages of the aforementioned types of analysis LPC "backward" and "forward", a technique of selectively associating the LPC analysis "back" and "forward" has been proposed in the article enti- lé: "Dual Rate Low Delay CELP Coding (8 kbits / s / 16 kbit / s) uslng has Mlxed Backward / Forward Rdaptlve LPC prediction" published by S.PROUST, C.LAMBLIN and D.MASSALOUX, Proc.IEEE Workshop Speech Cod. Telecomm. , Sept. 1995, pp 37-38. The conditions previously mentioned, concerning the proper operation of the LPC analysis "back", reveal that only this type of analysis has clear limitations when it is carried in transmission rates well below 16 kb / s. Besides the decrease in quality of the synthesized signal, which degrades the LPC filter performance, it is usually necessary to reduce the transmission rate, to operate on a frame length LPC greater, of the order of 10 to 30 ms. It is then found that in these conditions the degradation occurs primarily in the frequency spectrum of transitions and more generally in less stationary areas, while for generally very stationary signals such as those related to music, the LPC analysis "back "retains a very significant advantage vis-a-vis of the LPC analysis" before ".

The combination of the two above types of LPC analysis object is to overcome these disadvantages with the advantages inherent in each of them:

- LPC analysis "front" for the coding of transitions and non-stationary regions;

- Analysis "backward" LPC, higher order, for coding stationary areas.

In addition, the introduction of LPC frames coded LPC analysis "before" from the LPC coded analysis "backward" LPC frames allows the encoder and the decoder to converge back to the same synthesis signal in case of transmission error and therefore provides a very robust to these errors than a coding by analysis "backward" LPC pure.

Overall, the analysis mixed "forward" LPC - "back" above is to perform both analyzes LPC, an LPC analysis "before" the speech signal or audio encoding and an LPC analysis "back" on the synthesis signal.

Two filters are calculated for each block LPC, these filters are designated by LPC filter "forward" and "backward" LPC filter, respectively. A procedure for choosing the filter applied to the LPC block considered depending on the stationarity of the signal is implemented. This procedure uses two separate criteria:

- A first criterion based on filters prediction gains - a second criterion based on a distance parameter between LPC filters "before" calculated successively. For each of these two criteria, thresholds, fixed, are established.

First criterion: The selection of the "backward" LPC filter is retained if the difference between the prediction gain of the LPC filters "rear" and "front" is greater than a first threshold value.

Second criterion: For a current analysis of "backward" LPC analysis mode, prohibition of tipping LPC analysis mode "backward" LPC analysis mode "before" if the calculated distance on both filters representative parameter vectors "forward" LPC consecutive falls below a second threshold value, too little distance characterizing a substantially stationary area for which it is appropriate to avoid any change in LPC analysis mode. The calculated distance is a Euclidean distance between lès spectral lines of the speech signal or audio coding.

For a more detailed description of the aforementioned mixed LPC analysis mode, it is useful to refer to the article published by S.PROUST, C.LAMBLIN and D.MASSALOUX mentioned above.

Extensive investigations on the procedure of the aforementioned joint analysis helped to highlight the significant disadvantages below: - for some signals, the values ​​of the LPC filters prediction gains "front" and "rear" can oscillate on either side of the first threshold value. This phenomenon causes filter changes "backward" LPC - "forward" LPC, or vice versa, sudden and frequent. filtering the discontinuities then introduced an important source of deterioration of the synthesis signal and are not, for the most part, relate to actual spectral changes of the speech signal or audio to be coded; - The optimum value of the first threshold should be fixed, varies greatly depending on the stationarity of the signal to be coded, the more so that the coding rate is low. To a coding delay corresponding to an LPC frame of 10 to 30 ms, or when the transmission rate decreases, there occurs a sharp divergence of the coding mode of music and speech signals. For music signals, LPC analysis "rear" is almost continuously used while for the speech signal, the LPC analysis "before" is mainly used. While in the case of music signals, the stationarity is very high, the LPC analysis method "rear" is retained even for a frame length LPC important, in the case of speech signals, on the contrary, the very areas standing are very limited period and the transition to LPC analysis mode "back" in short consequence, causing unwanted transitions filters that degrade the quality of coding. The encoder is then no longer able to correct the phenomena generated by the discontinuity introduced by the tilting of the filters; - The LPC filter that gives the best subjective quality and therefore better models the spectrum of the signal to be coded is not always the one with the best prediction gain. Some failover from one mode of LPC analysis to another, related to an instant decision, are unnecessary.

The present invention aims to remedy the aforementioned drawbacks by implementing a method and a device for coding a digital audio signal by LPC analysis "front" and "rear" specific. Another object of the present invention is also the implementation of a dynamic adaptation of the choice function between the LPC analysis "forward" LPC analysis and "backward" according to the degree of stationarity of the signal coding. Another object of the present invention is also the implementation of a dynamic adaptation process the aforementioned choice function on the basis of discrimination between highly stationary signals, such as music or background noise and other signals, such as speech, to enable the most appropriate encoding processing by LPC analysis "back "and" before "respectively.

Another object of the present invention is also the choice of the most appropriate coding aforesaid having been made for a signal encoding a type or characteristics given ticks to prevent accidental tipping in the analysis mode LPC unsuccessful, and thus avoid the appearance of LPC filters transitions "before" - "back" or vice versa may reduce the quality of the reproduced signal synthesis. Another object of the present invention is finally the implementation of a dynamic adaptation process of the aforementioned selection function for which the analysis mode change LPC corresponds faithfully to a stationarity change of the signal to be coded and risk, the consequences of being less tied to a single crossing point effect of the first and second threshold values. The method and device for coding a digital audio signal, objects of the present invention implement a dual analysis on criterion of LPC analysis choice "front" and "rear" respectively for generating a transmitted coded signal consisting of LPC filtering parameters accompanied by an analysis decision information and an encoding residue signal, not transmitted. The digital audio signal is divided into frames, block sequence of a determined number of samples and the coding of the digital audio signal is performed on this signal from an LPC filter "front" for non-stationary zones and on a synthesis signal respectively, the synthesized signal being obtained from the encoding residue signal from an LPC filter "rear" for stationary areas. They are notable in that they consist in, and allow, respectively:

- Determining the degree of stationarity of the digital audio signal according to a tee stationnari- parameter, whose value is between a maximum stationarity value and a minimum stationarity value;

- Establish, from the stationarity parameter, an analysis choice value from a decision function; - Apply the value of the choice analysis filtering

LPC to perform encoding the digital audio signal by filtering "forward" LPC for non-stationary areas ners on the digital audio signal and filtering "backward" LPC for the stationary areas of the synthesis signal.

This procedure allows to favor the retention of a LPC filter modes "front" and "rear" respectively, in association with the degree of stationarity of the digital audio signal and to limit the number of switchings from one to other filtering modes and vice versa.

The method and the device, object of the present invention are applicable not only to the field of mobile telephony but also the creative industry and the reproduction of phonograms, transmission by satellite and high quality telephony video- applications or audio conferencing, multimedia.

They will be better understood from reading the description and studying the drawings below, in which:

- Figure 1 shows in the form of a general flow chart, an illustrative diagram of the steps for the implementation of the coding method, object of the present invention;

- Figure 2a shows a general flowchart calculation steps of the stationarity parameter for each current LPC block;

- Figure 2b shows a particularly advantageous embodiment of the essential steps of calculating the stationarity parameter according to Figure 2a;

- 2c shows a detail of the embodiment of Figure 2b, more particularly a detail of the refining value of the intermediate stationarity parameter process for obtaining the parameter stationnari- side;

- Figures 2d and 2e show a first respectively a second nonlimiting exemplary embodiment of implementation of a refining function to calculate a refining value of the intermediate stationarity parameter depending on the relative values ​​of the gain LPC filtering "front" and "rear" ^"; Figure 2f shows, by way of illustration, a flowchart of steps for the implementation of the decision function and value of the LPC analysis choice "front" or "back"; Figure 3 shows in the form of functional blocks, the general diagram of an encoder for performing encoding of an audio signal according to the object of the present invention; - Figure 4 shows in the form of functional blocks, the general diagram of a decoder for performing the decoding of an encoded audio signal through the implementation of an encoder such as shown in Figure 3. A description more detail the method of encoding a digital audio signal by dual analysis on choice criterion of LPC analysis "before" respectively "back" into a coded signal transmitted, object of the present invention, will now be given in conjunction with the figure 1.

Generally, it is indicated that the signal transmitted encoded, S_C. rated (t), is partly in the LPC filtering parameters accompanied by an LPC analysis decision information. In addition, a res coding residue signal_not(T) is transmitted not provided by the implementation of the coding method.

The digital audio signal is divided into LPC frames, LPC blocks succession, each block for the convenience of description, being denoted B_not and provided with a predetermined number N of samples. According to one aspect of the coding method, object of the present invention, it consists in carrying out the above encoding on the digital audio signal as previously defined from an LPC filter "front" for non-stationary zones respectively on a synthesis signal obtained from the encoding residue signal from an LPC filter "rear" for stationary areas.

According to a particularly notable aspect of the method of the present invention, it is to establish the filter selection criterion LPC "forward" or "back" on each current block of the succession of current blocks constituting a frame current, as shown in Figure 1, each current block, denoted B_notBeing available in an initial step 10, to determine in a step 11 the degree of stationarity of the digital audio signal according to a stationarity parameter, written STAT (n). This stationarity parameter has a numerical value between a maximum stationarity value, written STAT_MAnd a minimum stationarity value, written STAT_m.

By convention and, without diminishing the degree of generality of the object coding method of the present invention, it is indicated that the stationarity parameter provides the maximum value STAT_M for a strongly stationary signal, while this stationarity parameter provides the minimum value STAT_m for a signal highly non-stationary.

Following step 11 above, the object coding method of the present invention is to establish, in a step 12, from the stationarity parameter STAT (n), a value of LPC analysis choice, the choice value analysis corresponding course or at the choice of the LPC analysis "before" or rather the choice of the LPC analysis "back". The analysis choice value is denoted d_not(N) and is obtained from a function of specific decision, denoted D_not.

Step 12 above is then followed by a test step 13 for 1 application of value analysis choice_not(N), symbolized by C, the LPC filtering to perform encoding the digital audio-frequency signal by filtering "forward" LPC for non-stationary zones on the digital audio-frequency signal, respectively by filter "backward" LPC for stationary zones on the synthesis signal.

The implementation of the decision function D_not and the said analysis choice values_not(N), according to a particularly advantageous aspect of the coding method of the present invention, allows to favor the retention of a LPC filtering mode "forward" respectively "rear", in association with the degree of stationarity of audio-frequency signal, and to limit the number of switchings from one to another of the filtering modes, and vice versa.

Generally, it is indicated that the decision function implemented in step 12, the decision function is denoted D, is an adaptive function updated for each current block B_notFrom the stationarity parameter.

The updating of the adaptive function of maintaining focus in one of the LPC filtering modes "front", respectively "rear", depending on the degree of stationarity of the digital audio signal and thereby limiting the number ents toggle from one to the other of filtering modes, and vice versa.

More specifically, it is indicated that the value of analysis choice of_not(N) established from the decision function D_not above corresponds to a filtering mode priority value LPC "forward" or "backward" as well as another priority value representing in fact a value of no return priority to the LPC filtering mode "rear" or " before ". By filtering mode priority value LPC, it indicates that the value of choice analysis d_not(N) can correspond to a logical value, the true value of this logic value is 1 for example, matching a filtering option "backward" LPC while the value complementary to this true value is zero, corresponds to filter choice "forward" LPC. It is thus understood that the test function in step 13 can be reduced to a test case on the logical value of the value of the above analysis to ensure choice in step 14 the LPC filtering "rear" for zones stationary signal to be encoded or the LPC filtering "before" in step 15 for the non-stationary areas, the aforementioned steps 14 and 15 then being followed by the steps 14a and 15a return to the next block denoted B_{n + 1}, For n = n + 1. Although the value of analysis choice of_not(N) is represented by a logical value, we understand that this logic value may be associated with a value of priority and filter mode probability established by the decision function D_π specifically. in particular is understood that this probability value can correspond, for each current block B_notAt the true logic value for a range of probability values ​​between zero and one filter LPC "backward" while the logic value complemented, logical value zero for example, may correspond to the complement of the range above probability values ​​between zero and one of the first above range. This probability is related to the number of successive filtering decisions in the same filtering mode.

The procedure of the decision function D_not makes it possible to associate the logic variable_not(N) filtering mode priority value is adaptive over time, for each current block B_not.

Generally, it is indicated that the adaptation of the decision function D_not aims to gradually favor the LPC filtering mode "rear" or rather the LPC filtering mode "before" works best, given the overall stationarity of the signal to be coded, to avoid possible tipping useless one of the filtering modes to another. More specifically, it is indicated that:

- The stronger the signal to be coded is stationary and the decision function D_not emphasizes analysis "backward" LPC in minimizing switching to LPC analysis mode "before" - on the contrary, unless the signal to be coded is stationary and the decision function D_not favors the LPC analysis "before" in minimizing tipping LPC analysis mode "back".

A more detailed description of implementation of a specific decision function for performing the adaptation of the decision function, depending on the value of the stationarity parameter STAT (n) will be given later in the description.

A preferred method of calculation of the stationarity parameter STAT (n) on each current LPC block B_not will now be given and described in connection with Figure 2a.

According to the aforementioned figure, it is indicated that step 11 of determining the degree of stationarity of each current block B_not digital audio signal consists, from an arbitrary starting value of the parameter stationary, as shown in step 110 of Figure 2a, this arbitrary value being denoted STAT (O), calculated as a stage 111 for current block B_not a parameter value of the intermediate stationarity, denoted STAT * (n), according to a determined number of successive analysis choice values, these LPC analysis choice values, denoted d ^ Cn-l), ... to d_not._p(Np), being obtained for different successive blocks prior to the current block B_not the succession of LPC blocks, and the value of the stationarity parameter of the block preceding the current block, this value being denoted stationary STAT (nl). At step 111 shown in Figure 2a, it is indicated that the function of the determined number of previous analysis choice values ​​is given in relation to these previous values, denoted d_not.^ -l) To d_p__p(Np). Regarding the arbitrary starting value of the stationarity parameter STAT (O), it is indicated that it may, by way of nonlimiting example, be set equal to the average value between the maximum value and the minimum value of stationarity parameter mentioned above in the description, STAT_M and STAT_m.

The above step 111 is then followed by a step 112, which is to refine the value of the intermediate stationarity parameter based on the value of prediction gains filter or LPC analysis mode "front" and "rear" of the frame preceding the current frame. In step 112 of Figure 2a, it is indicated that the aforesaid function is denoted g (STAT * (n), Gpf, Gpb) where Gpf denotes the prediction gain of the LPC filter "front" and Gpb denotes the prediction gain LPC filter "rear" for frame preceding the current frame. At step 112, it is, at the end of step refine the parameter value of the intermediate stationarity, the stationarity parameter value STAT (n) of the current LPC block B_not assigned value, equation (3): STAT (n) = g (STAT * (n), Gpf, Gpb) corresponding to the refined value of the stationarity parameter intermediate rite.

A more detailed description of the calculation step 111 of the intermediate stationarity parameter STAT * (n) and step 112 consisting in refining the value of that parameter will now be given in conjunction with Figure 2b.

According to the aforementioned figure, step 111 is, from an initialization step 1110 in which the value of the stationarity parameter STAT (n) and the value of choice analysis d_not-. (Nl) related to the LPC block B_not.₁ prior to the current block B_not is available, to operate in a step 1111, a step of discriminating LPC analysis mode "forward" or "backward" LPC block B_not,₁ preceding the current block B_not. This step can discriminate in 1111, as shown in Figure 2b, consist of a test step on the value of choice analysis d_not._:(Nl) to the symbolic value "fwd" or logic zero value corresponding to the complemented value of the true logic value.

On a negative response to the test in 1111, cited above, that is, for every block B ^ preceding the current LPC block B_not analyzed in "backward" LPC analysis mode, the step of calculating the parameter value of the intermediate stationarity consists, in a step 1113, to determine the number of previous frames consecutively analyzed by LPC analysis mode "rear" N_BWD number noted then, in a step 1114, compared to superiority comparison criterion the number of previous frames to a first arbitrary value, written Na, representative of a number of successive frames analyzed LPC Mode "back". On a positive response to the superiority comparison of the 1114 test, the calculating step then is to assign, in a step 1114b, the value of intermediate stationarity parameter STAT * (n), the value of the stationarity parameter of the block preceding the current block, STAT (nl), increased by a determined value based on the first representative value of a arbitrary number of frames successive analyzed, that is to say in fact the number of previous frames consecutively analyzed in N BWD LPC analysis mode "rear". In step 1114b, the specific function of the first arbitrary value value is denoted f_at(N_BWD). In the aforementioned step, we understand that the value of the intermediate stationarity parameter STAT * (n) for the current LPC block B_not is increased relative to the corresponding value of the same stationarity parameter for the previous block B_not.₁. On a negative response to the superiority comparison to 1114 comparison test, the value of intermediate stationarity parameter STAT * (n) is assigned, in a step 1114a, the value of the stationarity parameter STAT (n-1) of the preceding block the current block B_r . In contrast, for any previous block B_not.₁ analyzed by LPC analysis mode "before", that is to say, on positive response to test 1111, the step of calculating the intermediate stationarity parameter 111 is, as shown in Figure 2b, to determine in a step 1112, on test criterion the occurrence of a transition from "backward" LPC analysis mode LPC analysis mode "before" between the block prior to the previous block in current block B_not__lr of rank n-2, that is to say, the existence of a value of LPC analysis choice_not.₂(N-2) = symbolic value "bwd" or logic zero as mentioned previously. The positive response to the test 1112 indicates the existence of such a transition "back" mode of analysis for the LPC block B_not_₂ preceding the block preceding the current block _not,_lf while a negative response to the said test 1112 indicates the absence of such a transition.

Upon a positive response 1112 supra occurrence test, the calculation step 111 is then to compare, on criterion inferiority comparison, the number of previous frames N_B aforementioned D to a second arbitrary value representative of a number of N_ successive frames analyzed LPC Mode "rear" preceding the block B - ^ ,. preceding the current block.

On a positive response to the comparison made in the 1118 test, this test is followed by a step 1118a of assigning to the parameter value of stationary intermediate STAT * (n) the value of the stationarity parameter of the block preceding the current block STAT (n) minus a predetermined value, according to the second arbitrary value N_bThis determined value is denoted f_b(N_BWD). It is thus understood that in the step 1118a award, the value of the intermediate stationarity parameter is thus reduced accordingly.

On the contrary, a negative response to the inferiority comparison test conducted in 1118, then step 111 is to assign, in a step 1118b, the value of the stationarity parameter via STAT * (n) the parameter value of stationarity précédant- block of the current block, ie STAT (nl).

In Figure 2b, note that the steps 1118a and 1118b attribution are then followed by a reset step of the number of successive blocks treated as "backward" LPC analysis mode, the reset stage wearing the reference 1118c and to update all the calculation process of the value of the intermediate stationarity parameter. On a negative response to the comparison test 1112, no transition LPC analysis "before" being appeared, the value of the intermediate stationarity parameter STAT * (n) is assigned the value of the stationarity parameter STAT (n) of the previous block B ^ in a step 1119. at the end of step 111, it has the value of the intermediate stationarity parameter STAT * (n) for the current block B_not.

Regarding step 112 comprising refine the parameter value of said intermediate stationarity, it is indicated with reference to Figure 2b, as the latter can advantageously consist, in a step 1120, to discriminate against earnings prediction LPC filtering "rear" and the filter "forward" LPC, these gain values ​​being marked Gpb and Gpf respectively. It is understood that the aforementioned discrimination step is to simply memorize and read the gain values ​​calculated for the LPC filter "before" respectively "rear" supra. Apart from the above gain values, step 1120 may be to calculate the relative value of prediction gains, noted DGFB, such as the difference or ratio between earnings prediction "front" and "back" above.

As was further shown in Figure 2b, step 112 of Figure 2a comprises a result of the above step 1120 a step 1121 consisting in modifying the value of the intermediate stationarity parameter STAT^*(N) a refining value .DELTA.S, this refining value in accordance with a particularly notable feature of the method of the present invention is a function of the relative value of LPC filtering prediction gains "front" and "rear" . In general, it is indicated that the function representative of the .DELTA.S refining value is denoted:

- .DELTA.S = F_r(Gpf, Gpb) where Gpf and Gpb designate as previously LPC filtering prediction gains "before" respectively "rear". Generally, it is indicated that the function f_r(Gpf, Gpb) for establishing the refining value .DELTA.S is decreasing respectively increasing function of this relative value, depending on the direction in which the relative value considering. When the relative value is the value of the "backward" LPC filtering gain relative to the filter gain "forward" LPC, this choice may be arbitrarily detained, without diminishing the generality of the method according to the invention, relative value DGFB prcitée, the function f ,. is then increasing. It decreases in the opposite case.

In other words, the modification, increase by tion or decrease in the value of the intermediate stationarity parameter .DELTA.S refining value is proportional to the relative value of gains. Overall, this modification is written STAT (n) STAT^*(N) + kΔS. In practice it will k = 1. More specifically, it is indicated that the refining value .DELTA.S increases in algebraic value when 1 difference between the LPC filtering prediction gains "front" and "rear" increases, function f_r(Gpf, Gpb) then being an increasing function, whereas this .DELTA.S refining value decreases in algebraic value when the same aforementioned gap decreases, the aforementioned gap is defined between the prediction gain of the LPC filtering "rear" and the gain prediction filtering "forward" LPC. In fact, this function is increasing or decreasing according to the definition of this gap.

As a result, the end of step 1121 as shown in Figure 2b, the value of the intermediate stationarity parameter STAT^*(N) can then, for k = 1, be corrected by the algebraic value of the said refining value .DELTA.S to calculate the value of the stationarity parameter STAT (n).

Following step 1121, and it has the value of the stationarity parameter STAT (n) in step 1122.

A more detailed description of step 1121 of Figure 2b will now be given in connection with Figure 2c in a preferred embodiment wherein a plurality of testing criteria are applied to both the refining value to the values ​​of LPC prediction gain "front" and "rear" to optimize the process of calculating the stationarity parameter.

As shown in the aforementioned FIG 2c, step 1121 may consist of a first step 1121a to calculate the refining value .DELTA.S from the function f_r(Gpf, Gpb) cited above. Various examples of usable functions will be given later in the description. First, the .DELTA.S refining value is subject to a superiority comparison test to the value 0, in a step 1121b, the comparison test to actually determine the increase of the refining value .DELTA.S. On a positive response to the aforementioned test 1121b, the refining value .DELTA.S being positive and corresponding to an increase in the relative value of the LPC filtering prediction gains "front" and "rear", the step of increasing the value of intermediate stationarity parameter value the .DELTA.S refining is further subjected to a condition of superiority of filtering gain value "backward" LPC relative to a first determined positive value, in a step for comparing superiority of the value of the filter gain "backward" LPC Gpb relative to said first determined positive value, denoted S.

On a negative response to the aforementioned test 1121c, the value of the stationarity parameter STAT (n) is assigned the value of the intermediate stationarity parameter STAT^*(N) in a step 1121g. On a positive response to the aforementioned test 1121c, increasing the value of the intermediate stationarity parameter value the .DELTA.S refining is further subjected to a condition of inferiority of the value of the intermediate stationarity parameter STAT^*(N) to a second determined positive value STA ^ course representing a stationary value. This inferiority condition test is carried out 1121st step.

On a negative response to the test 1121st cited above, the value of the intermediate stationarity parameter STAT (n) is assigned the value of the intermediate stationarity parameter STAT^*(N) at the above step 1121g.

On a positive response to the inferiority of condition test 1121st the value of the intermediate stationarity parameter STAT (n) is assigned the value of the intermediate stationarity parameter STAT^*(N) plus the .DELTA.S positive value of the refining value in step 1121i. Conversely, on negative response to the aforementioned test 1121b, the .DELTA.S refining value is negative, the reduction step of the intermediate stationarity parameter value of the refining .DELTA.S, this value being negative, is further subjected to a test of inferiority condition of the filtering gain value "backward" LPC Gpb against a determined third positive value denoted S_d 1121d in a comparison step. This third determined positive value is obviously representative of a filtering gain value LPC.

On a negative response to the test 1121d supra stationarity parameter value STAT (n) is assigned the value of the intermediate stationarity parameter STAT^*(N) in step 1121g. On the contrary, on a positive response to the aforementioned test 1121d, the step of decreasing the value of the intermediate stationarity parameter .DELTA.S refining value is also subject to a superiority condition of the value of the intermediate stationarity parameter STAT^*(N) relative to a fourth specified positive value, denoted Statd in a comparison test noted 1121f. Of course, the fourth determined positive value represents a stationarity parameter value selected.

On a negative response to the test 1121f supra, at stationary STAT parameter value (n) is assigned the value of the intermediate stationarity parameter STAT^*(N) in step 1121g.

On a positive response to the test 1121f supra, at stationary STAT parameter value (n) is assigned the value of the intermediate stationnaritée parameter STAT^* increased the algebraic value of the refining value .DELTA.S, negative, the value of the intermediate stationarity parameter being thus reduced to establish the stationarity parameter value STAT (n) in step 1121h. In late stages 1121g, 1121h and 1121i, so one has to step 1122 of Figure 2b of the parameter stationary STAT (n).

As regards the function f_r(Gpf, Gpb), indicate that the latter may consist of a nonlinear function of the relative value of LPC filtering gain "front" and "rear" in which the relative value of LPC filtering prediction gains "before" and "rear" may itself be either in the report or by the difference of the LPC filtering prediction gains "front" and "rear". Other types of functions, such as linear functions, may be used.

A first example of a non-linear function f_r(Gpf, Gpb) is shown in Figure 2d.

In the embodiment of Figure 2d, pairs of value of the prediction gain of the LPC filtering "rear" Gpb plotted as ordinate and LPC filtering gain "before" Gpf used to assign positive values ​​.DELTA.S refining , .DELTA.S> 0 or negative .DELTA.S <0 for a value of the ratio p = Gpb / Gpf corresponding to a slope greater respectively smaller than the right side .DELTA.S = 0. in Figure 2e, the case is shown where the relative value of filtering prediction gains LPC "front" and "rear" no longer corresponds to earnings ratio p but unlike the aforementioned gains.

In this case, the function of the relative value of the LPC filter the prediction gains of "forward" and "rear" f_r(Gpf, Gpb) may also be a nonlinear function for assigning the .DELTA.S refining value for values ​​of this difference corresponds to pairs of Gpb value, corresponding to straight Gpf whose x-intercept is respectively less than, in algebraic value with the abscissa at the origin of the right .DELTA.S = 0.

In the case of Figure 2, lines delimiting the zones according to the sign of the .DELTA.S refining value are mutually parallel.

According to another particular aspect of method of the invention, it is further states that it should not fit 1 'stationary index of the current block B_not during silence frames, for example when the audio signal comprises a speech signal having silences. In such a case, step 1111 of step 111 shown in Figure 2b may be preceded by a step 1111a, for each successive current block, to determine the average energy of the digital audio signal and to compare in the same step on criterion inferiority comparison, the average energy in a determined threshold value representative of a silence frame. In Figure 2b, this threshold value is denoted ENER_SIL. On a positive response to the above test, the value of the stationarity parameter STAT current block (n) is assigned the value of the stationarity parameter of the previous block STAT (nl) to 1111b assigning step shown in Figure 2b. Steps 1111a and 1111b are in the aforementioned figure, shown dotted, as Reserved for example the coding of a speech signal. A more detailed description of the implementation of the decision function D_not allowing the production of d_ decision values ​​(n) will now be given in conjunction with FIG 2f. This description is given in a preferred embodiment in which this decision function, which can be compared to that described in the article previously mentioned in the description, published by S.PROUST, C.LAMBLIN and D.MASSALOUX is however temporally adapted, in accordance with the object of the present invention in order to obtain the values ​​of analysis choice of d (n) successive.

From a step 120, for the current block B_notIs calculated first distance, d_LPCBetween the LPC filter of the current block and the previous block _not^. This distance calculation is done for example by using the LSP frequency parameters as mentioned above in the description with respect to the method described in the aforementioned article. We notice :

- S_PRED (n) and S_TRANS, S_STAT and G_x the values ​​of the thresholds involved in the test based on the LPC filter prediction gains "back" and "forward";

- S_LSP_L S_LSP_H and the values ​​of thresholds involved in the criterion based on the distance between LSP frequency vectors representing two LPC filters "before" relating to two consecutive blocks B_not._l and B_not ; - Gpf the prediction gain of the LPC filter "front";

- Gpb the filter prediction gain "rear"; and

- Bcf the filter prediction gain "before" interpolated using the method described in the article mentioned above in the description. The criterion for establishing the function of decision in relation 2f, is made as described below:

- If the consecutive LPC filters are stationary, that is to say for d_LPC <S_LSP_L then no failover filtering "backward" LPC filtering "forward" LPC is achieved if one is filtering mode "backward" LPC, provided that the prediction gain of the LPC filter "rear" is greater than the prediction gain of the LPC filter "front" minus a S_STAT value. It indicates that the S_STAT value is chosen to promote the choice of a LPC "rear" filter in the presence of a large stationary the measured spectrum to 1 using the distance_LPC ;

- If the consecutive LPC filters have a major transition, that is to say for d_LPC > S_LSP_H and if Gpf> Gpb-S_TRANS, then the filtering mode selected is the LPC filter "before", that is to say of_not(N) = 0, symbolic value "fwd" otherwise, of_not(N) is set equal to 1, symbolic value "bwd". It is indicated that the value of S_TRANS is selected so as to strongly favor the choice of the LPC filter "front" in the presence of a spectral transition measured 1 using the distance d_LPC ; Otherwise, in all other cases, if Gpb> Gpf-S PRED and Bcf> Gpf-S_PRED then, the successful LPC filter is the LPC filter "back" interpolated, provided that the gain of the latter and that of the LPC filter "backward" pure exceed the threshold value G_: previously mentioned. If the condition on the above prediction gain values ​​is not met, then the LPC filtering is selected "before".

In order to increase the number of LPC filters "before" transmitted, thereby increasing the robustness of the coding system to transmission errors, the LPC filtering mode "forward" can be advantageously selected as soon as the energy of the signal to be coded E_notThat is to say, the energy of the block B_not correspondent, becomes less than the value of the energy of a silence frame ENER_SIL, this energy value corresponding to the minimum audible level.

All the conditions for the establishment of the decision function D_r and obtaining analysis choice values_not(N) corresponding, is illustrated in Figure 2f with temporal adaptation of the decision function D_not. The value of the stationarity parameter STAT (n) may for example be marked on a scale of 0, corresponding to the value STAT ... very little standing, 100, corresponding to the value STAT_M very steady. Depending on the value of the stationarity parameter

STAT (n), the decision function D_not is modified by adjusting the threshold values.

Over the stationarity of the signal increases, the filtering mode "backward" LPC is favored: the S_PRED, S_LSP_L and S_LSP_H thresholds are increased.

By way of nonlimiting example, the editing functions are indicated for each current LPC boc B_not the above threshold values:

- S_PRED (n) = f_{s? sε} (STAT (n)) with f_{s PRED} increasing function of the value of STAT (n);

- S_LSP_L (n) = f_{s LF:} . (STAT (n)) with f_{The LPC s} function increasing;

- S_LSP_L (n) = f_{LPC s H}(STAT (n)) with f_{s LPC} "Increasing function.

In the adaptation of the above threshold values, it is indicated that the increasing functions mentioned are, for example step functions with regard to the functions f_{The LPC s} and F_{s H CPA}. The function f_{s PRED} is an affine function of the variable stationarity parameter, of the form: S_PRED (n) = α.STAT (n) + β where a and β are two real values ​​between 0 and 1 and where the value of S_PRED (n) is bounded in the interval [S_PRED_m, S__PRED_M] S_PRED-- and S_PRED_M represent two experimentally determined values. To further reduce the risk of tilting of the filters, it is then possible to choose, where the stationarity parameter STAT (n) is less than a threshold value S_{r ;:} given, to impose filtering mode "forward" LPC. By against the threshold values ​​S_TRANS, S_STAT and

G_x maintain a fixed value, these values ​​being, for example equal to -1 dB, 5 dB and 0 dB, respectively.

The establishment of the decision function D_not and obtaining analysis choice values_r(N) are illustrated as follows in Figure 2f: following the aforementioned step 120, performing a test step 121 on the energy of the current LPC block B_not, By an inferiority comparison to the silence ENER_SIL energy value or the value of the stationarity parameter STAT (n), compared by an inferiority comparison to the value S_FUD cited above in the description. On a positive response to test 121 cited above, the value of choice analysis d_not(N) is set equal to 0, ie symbolic value "fwd" in step 122. On a negative response to test 121 cited above, a new test is performed relative to the value of choice of analog lysis of_not__x(Nl) to logic 1, that is to say, the symbolic value "bwd".

On a positive response to the aforementioned test 123, a new test is performed on the filtering distance d LPC_LPC supra, in a step 124, compared with the threshold value

S_LSP_H (n) by comparing superiority to this threshold value.

On a positive response to test 124 cited above, a new 126 a test is performed, consisting of comparing the prediction gain of the LPC filtering "forward", Gpf, the prediction gain of the LPC filtering "rear" Gpb minus the threshold value S_TRANS.

On a positive response to the aforementioned test 126a, the value of choice analysis d_not(N), is assigned the logical value 0, symbolic value "fwd" and a negative response to the aforementioned test 126a, is attributed to the same value of choice analysis logic 1, symbolic value "bwd". The corresponding steps are denoted 128 and 129.

On a negative response to test 124 mentioned above, a new test 125 is performed. The test 125 is to perform a comparison of the distance of LPC filtering, d_LPC, By inferiority comparison to the threshold value S_LSP_L (n).

On positive response to test 125, a new 126b test is performed by comparing superiority of filtering prediction gain "backward" LPC filtering prediction gain "forward" LPC S_STAT minus the value cited above.

On a positive response to test 126b, the value of analysis choice of_not(N) is assigned in step 129 the logical value 1, that is to say, the symbolic value "bwd".

On a negative response to test 126b, the value of analysis choice of_not(N) is assigned the logical value 0, that is to say, the symbolic value "fwd", step 128. Conversely, on negative response to test 125, a new test is performed, in a step 127, the test consist as to check the comparison condition of the LPC filtering gain "backward" Gpb the filtering prediction gain "forward" LPC decreased by the threshold value S_PRED (n), comparing superiority of filtering prediction gain LPC intermediate Bcf the value of filtering prediction gain "forward" LPC minus the abovementioned threshold value S_PRED (n) and comparing superiority of filtering prediction gain "backward" Gpb to the threshold value G_x And comparing the value of the intermediate filtering prediction gain Bcf to the threshold value G_x .

It indicates that the negative response to test 123 previously mentioned in the description also leads to the realization of the aforementioned test 127. On a positive response to test 127 previously mentioned, the value of choice analysis d_not(N) is assigned the logical value 1, that is to say, the symbolic value "bwd" in step 129, whereas the negative response to test 127 cited above, the value of choice analysis_not(N) is instead attributed the logical value 0, that is to say, the symbolic value "fwd" in step 128.

It provides, through the implementation of the decision function D_r, The value of analysis choice of_not(N) obtained with the logical values ​​1 or 0 above, these logical values ​​being however associated to a priority value or no return priority filtering mode "backward" or "forward" according to the value of the parameter stationarity.

A more detailed description of a device for encoding a digital audio signal by dual analysis of specific LPC analysis choice "front" respectively "back" into a coded signal transmitted, according to the object of the present invention, will now be given in connection with Figure 3. in a practical manner, it is indicated that the digital signal to be coded is divided into frames consisting successive blocks of samples, each block comprising a given number N of samples, for example.

In Figure 3, the method of granting a digital audio signal to be encoded into successive blocks of samples B_not not shown because this procedure is well known in the art and can be made from a simple buffer, for example addressed by reading periodically the frame frequency and the frequency block. As further shown in Figure 3 above, the object coding device of the invention comprises an LPC analysis filter "before", carrying the reference 1A, and an LPC analysis filter "back", designated 1B, to enable to grant a transmitted encoded signal consisting of LPC filtering parameters accompanied by an analysis decision indication of, as well as parameters Pr_r related to harmonic analysis and the CELP excitation signal.

Generally, it is indicated that the indication of decision analysis corresponds to the value of selection of analysis_rι(N) as mentioned above in the description. Regarding the LPC filtering parameters, it is noted that these correspond to specific parameters, in accordance with in embodiment of the object coding method of the present invention, as will be described below in the Description.

In Figure 3, there is also shown, in the coding device according to the invention, the existence of an adaptive filter based on the value of the stationarity parameter, adaptive filter carrying the reference 1E. This adaptive filter 1E receives of course the original digital signal, denoted s_{n (t)}, That is to say the current block B_not. 1E filter uses the LPC filtering parameters to calculate the residual signal which will then be encoded by the IF module. These LPC parameters, as well as indication of filtering decision is part of the coded signal which is transmitted to the decoder.

In addition, as shown in Figure 3, the object coding device of the present invention comprises a coding means, bearing the reference IF, a non-transmitted coding residue signal, the encoding residue signal, designated by res_{n (t)} is directly available at the output of the adaptive filter 1E, which signal is thus delivered to the input with the digital audio signal to the encoding module untransmitted encoding residue signal, for generating a synthesis residue signal, res_syn_not(T).

Inverse filtering module, bearing the reference 1G, receiving the synthesis residue signal and allows to grant a synthesis signal referenced s_syn_{n (t)}.

A 1H memory module receives the aforementioned synthesis signal s_syn_{n (c}To deliver the aforementioned synthesis signal for the previous block to the current block B_not, The synthesis signal thus obtained being denoted by s_syn_not.₁(T). This synthesis signal is delivered to the LPC analysis filter "back" carrying the reference 1B in Figure 3 above. The coding device, object of the present invention, as shown in Figure 3, allows to perform coding of the digital audio signal on the audio aforementioned digital signal from the LPC filter "front" for non-stationary zones and on the aforementioned synthesis signal s_syn_not.₁(T) from the LPC filter "rear" 1B for stationary zones, as will be described below.

As will be observed in Figure 3 above, the object of the invention device includes for this purpose, for each current LPC block B_not, 1C a module for calculating the degree of stationarity of the digital audio signal according to a stationarity parameter the value of which is between a maximum stationarity value and a minimum stationarity value. Of course, the stationarity parameter is the parameter STAT (n) described above in the description according to the object coding method of the present invention. The maximum and minimum stationarity values ​​are also defined above. As was further shown in Figure 3, the coding device according to the invention comprises a module, denoted 1D_X, Establishment from the aforementioned stationarity parameter STAT (n) of a decision function and a value of LPC choice analysis, the decision function is denoted D_not as previously mentioned in the description and analysis choice value being LPC course and corresponding to the value of LPC analysis choice denoted d_not(N) described above in the description. Remember that the value of choice analysis d_not(N) can take the values ​​0 or 1, logical values, which correspond to the symbolic value of choice analysis "fwd" and "bwd" for the LPC analysis "front" and "rear" respectively.

is understood in particular that with regard to the means of establishing the decision function D_not, The latter corresponds to a software implementation, for example, as described above in connection with FIG 2f. Furthermore, the coding device according to the invention as shown in Figure 3 comprises filter analysis discrimination module LPC noted 1D₂This module receiving the value of choice analysis d_not(N) and for delivering to the current LPC block B_not the value of the LPC filtering parameters "rear" respectively "front" depending on the value of selection of said analysis. We understand of course that the filtering analysis parameters "backward" LPC analysis and the LPC filtering parameters "before" are of course available in digital form at the filter carrying the reference 1B and 1A respectively in Figure 3. These parameters are designated respectively Af_not(Z) for filtering the scan settings "forward" LPC regarding the LPC analysis filter "front", carrying the reference 1A, and Ab_r(Z) the LPC analysis parameters "rear" with respect to the LPC analysis filter "Rear" carrying the reference 1B. These parameters are issued 1D module. and 1D Module₂ respectively.

As regards the hardware realization of discrimination module 1D₂, It is indicated that it may, for example, in a non-limiting embodiment, consist of two separate memory areas for storing the Af filter settings_not(Z) and Ab_not(Z) respectively, the value of analysis choice of_not(N) based on its current logic value 0 or 1, to 1 'read addressing filtering parameter values ​​stored by the unit 1D₂ for example, and the transmission of these filter parameters by the latter.

Finally, as shown in Figure 3, it is indicated that the encoding device according to the object of the present invention, for carrying out the adaptive filter as a function of the stationarity value carrying the reference 1E, can be achieved by a filter element whose transfer function, denoted A (z), is established from the filtering parameter values ​​delivered by the discrimination module 1D₂ previously mentioned.

It is thus understood that the adaptive filtering module 1E can be achieved by adjustable coefficients a filter, the coefficients of the latter being assigned the filtering parameter values ​​delivered by the discrimination module 1D₂ previously mentioned. The filtering carried out by the module 1E and adaptive type according to the degree of stationarity of the digital audio signal to be encoded. 1E module and delivers, from the digital audio signal original s_{n (t)}, The filtering residue signal LPC designated by res_not(T) to the encoding module IF residue, which then is used to deliver the LPC synthesis residue signal designated by res_syn_not(T).

Finally, the module 1G is a filter module whose transfer function is 1 the inverse of the transfer function modulus 1E obtained pa RTIR parameters stored latter. He receives the synthesis residue signal LPC res_syn_not(T) delivered by the coding module coding residue delivered by the IF module. It is thus understood that encoding the digital audio signal s_not(T) is performed at the 1E module using the LPC analysis "front", respectively "rear" performed by the LPC analysis filter "forward" LPC analysis 1A and "rear" 1B, the encoded signal S_C_not(T) consisting of the transmission of the LPC filtering parameters "forward" when the value of selection of analysis_not(N) has the symbolic value "fiv'd" and the indication of the choice analysis, that is to say the value of choosing aforementioned analysis. This procedure allows for encoding the digital audio signal and prioritize maintenance in one of the LPC filtering modes "front", respectively "rear", depending on the degree of stationarity of the digital signal and limit further the number of switchings from one to another of the filtering modes considered.

A device for decoding a digital audio signal encoded by dual analysis of specific LPC analysis choice "front", respectively "rear", into an encoded signal transmitted in accordance with the object coding method of the present invention, and with the implementation of an encoding device as shown in Figure 3 for example, will now be described in connection with Figure 4.

In general, it is indicated that the transmitted coded signal S_C_not(T) consists, for each analysis block

LPC in the selection value of above assay and, in the case where the value of choice analysis corresponds to the LPC analysis block considered to a LPC analysis "front", in the LPC filtering parameters "before" and the encoding parameters of the filter residue LPC parameters Pr_not, That is to say the signal res_not(T) into a res_syn synthesis residue signal. (T) by the encoding module of the IF residue. As shown in Figure 4, it is indicated that the decoding device comprises at least one module synthesis, referenced 2A, the filtering residue signal receiving the coding parameters of the LPC residue output by the IF module. The module 2A decodes the encoding provided by the IF module parameters and consequently delivers a synthesis residue signal, which is referenced in Figure 4 res_syn_not(T).

The decoding device as shown in Figure 4 also includes a module carrying the reference 2B, adaptive inverse filtering according to the degree of stationarity, receiving the synthesis residue signal previously mentioned, provided by module 2A, and enabling 'create a synthesis signal s_syn_not(T) representative of the digital audio signal, this signal in fact constituting the decoded signal. We understand of course that the reverse filtering module 2B uses the filtering parameters received by the decoder because of the transmission or the LPC analysis parameters "before" when they are transmitted and that the decision to analysis corresponds to an LPC analysis "forward" or, on the contrary, filtering analysis parameters "rear" as will be described below.

For this purpose, the decoding device of the present invention of course comprises a filter module "backward" LPC, bearing the 2D reference, receiving the synthetic signal, ie the signal referenced s_syn_not(T) for the previous LPC block the current LPC block, synthesis signal thus being referenced s_syn_not.₁(T) Figure 4. It includes for this purpose that the synthesis signal relative to the current block B_r and referenced s_syn_not(T) can then be issued to the filter module "backward" LPC 2D via a storage module, bearing the reference 2E in fact possible, for an addressing in appropriate reading, shifting the reading of the signal synthesis that corresponding to the block preceding the current block B_not. Finally, to ensure the above procedure, the object decoding device of the present invention, as shown in Figure 4, also includes a discriminator module bearing the reference 2C, for performing a discrimination of the LPC analysis "front", respectively "rear". 2C module receives, firstly, to control of discrimination, the choice analysis value received, that is to say the value of_not(N), and, on the other hand, LPC filtering parameters "forward", that is to say the parameters Af_not(Z) transmitted, and the LPC filtering parameters "rear" Ab_not(Z) obtained by the 2D module. 2C module and allows delivery, depending on the value of analysis choice, that is to say the value of_not(N) or the LPC filtering parameters "before" Af_not(Z) or filter settings "backward" LPC Ab_not(Z) to the adaptive filtering module reverse 2B according to the degree of stationarity.

As regards the hardware embodiments 2C and 2B modules, it is indicated that these can simply consist of substantially identical modules modules 1D₂ and 1E, or more particularly, 1G of Figure 3.

As regards the actual implementation of a coding device according to the object of the present invention, enabling implementation of the method as described previously in the description, two specific embodiments have been made.

* CELP coder telephone band, in high-speed expansion of the ITU-T standard to 8 kb / s:

The encoder itself consisted of an encoder in the telephone band of 300 to 3400 Hz, at a rate of 12 kb / s CELP. The frames were made over a period of 10 ms for an excitation provided by algebraic dictionary using the technique called ACELP previously mentioned in the description. The LPC analysis "before" was an order analysis and analysis LPC 10 "rear" an order analysis 30 all 80 samples.

A separator for the encoding of the residue into two sub-blocks of 40 samples was conducted. Each block B_not contained 80 samples.

Adaptation of the stationarity parameter STAT (n)

The aforementioned stationarity parameter varies between two extreme values ​​0 and 100, the values ​​STAT_m and STAT_M above.

The functions of adjustment described above in the description, and in particular the functions f_at(N_BWD) and f_b(N_B D) were such that:

1, 56 if N_BWD> 20 f_at(BWD 7.81 N = 20 if N_BWD 0 otherwise

I 0.78. (20-N_BWD) if f ≤ 20 N_BWD_b(N_BWD) i = 0 otherwise. In these relationships, x = DGFB.

As regards the function f_r for establishing the value .DELTA.S refining mentioned previously in the description, this is a step function of the variable x, where x = Gpb - Gpf and .DELTA.S = f_r(X) and whose value is:

<img id="imgf000040_0001" he="71" wi="67" file="imgf000040_0001.tif" img-format="tif" img-content="drawing" orientation="portrait" inline="no" /> The refining STAT (n) is further subject to the following conditions mentioned above with 2c:

If .DELTA.S&gt; 0:

If STAT * (n) &lt;stati STAT (n) = STAT * (n) + .DELTA.S Otherwise STAT (n) = STAT * (n)

If not

STAT (n) = STAT * (n) = 40.6 with stati.

The other test conditions referenced 1121d,

1121c and 1121f Figure 2c were not used in this embodiment.

Adaptation of decision thresholds As regards the decision thresholds:

S_PRED is modified as follows:

S_PRED (n) = 0,03.STAT (n) + 1.0

S_PRED € [S_PRED_m, S_PRED_M] = 1.03 and S_PRED S_PRED_M = 4;

The S_LSP_L threshold is suitable for 1 using the function by following steps: j 0,015 if STAT (n) = 100

S_LPS_L (n) = f_s__L3F_._ (STAT (n)) = lO otherwise

The value of S_STAT threshold used when stationary LPC filters measured using S_LSP_L threshold was set to 4.0 dB.

S_LSP_H the threshold has not been used in this embodiment.

The threshold value G_λ was set at OdB. As regards the energy value characterizing a ENER_SIL silence frame, this value was set at 40 dB measured on 80 samples s (i) of the current block B_not :

ENER 10.log SIL = \ L s (i Regarding the value of Spy- threshold mentioned above and intended to further limit the tilting risk by requiring the LPC filtering mode "before" when the value STAT (n) is below this threshold, the value S_FUD was set at 40.6.

* A second embodiment of a CELP encoder wideband two sub-bands 16/24/32 kb / s was performed under the following conditions:

- Wideband encoder from 0 Hz to 7000 Hz into two subbands. A main band was coded with CELP, frame of 120 samples, excitement created by algebraic dictionaries, and transmission of certain characteristics and energy spectrum of a host band between 6000 Hz and 7000 Hz. - LPC analysis " panel "14 coefficients and analysis

"Backward" LPC coefficients 50 every 120 samples.

LPC analysis mode "before" separation into two sub-blocks

LPC 60 samples, the filter used for the first sub-block is interpolated from the current filter and the previous filter.

Calculation of the stationarity parameter STAT (n)

In this embodiment, the aforementioned stationarity parameter varies between two extreme values ​​0 and 120, STAT values_m and STAT_M above. As regards the adaptation of the value of the stationarity parameter STAT (n), the values ​​of the functions f_at(N_BWD) and f_b(N_BWD) are such that:

4 if N_BWD&gt; 10 f_at(N_BWD) = 20 = 10 if N_BWD

0 otherwise

(10-N_BWD if N_BWD &lt;10 f_b(N_BWDι = 10 otherwise.

As regards the function f_r allowing estab- lish the .DELTA.S refining value mentioned previously in the description, this is a step function of the variable x, where x = Gpb / Gpf and .DELTA.S = f_r(X) and whose value is:

[

[

<img id="imgf000043_0001" he="65" wi="72" file="imgf000043_0001.tif" img-format="tif" img-content="drawing" orientation="portrait" inline="no" />

The refining STAT (n) is further subject to the following conditions mentioned above with 2c: If .DELTA.S&gt; 0: <img id="imgf000043_0002" he="5" wi="28" file="imgf000043_0002.tif" img-format="tif" img-content="drawing" orientation="portrait" inline="no" />

If STAT * (n) &lt;STAT_; STAT (n) = STAT * (n) + .DELTA.S Otherwise STAT (n) = STAT * (n)

Otherwise STAT (n) = STAT * (n)

If not :

If STAT * (n) &lt;stati STAT (n) STAT * (n) + .DELTA.S Otherwise STAT (n) STAT * (n) with stati = 80, if = OdB.

The other test conditions referenced 1121h and

1121d of Figure 2c were not used in this embodiment.

Adaptation of decision thresholds As regards the decision thresholds: S_PRED is adapted as follows:

S PRED (n) = 0,03.STAT (n) - 0.5 bounded in the interval [S_PRED_m, S_PRED_M], With S_PRED_m = 0.5 and S_PRED_M ≈ 2.5.

The S_LSP_L threshold is suitable for 1 IDE in the following step function: 5 j0,02 if STAT (n)&gt; 100

S_LPS_L (n) = f_s__LSP__L(STAT (n)) = Iq, 01 otherwise

S_LSP_H the threshold is adapted by using the following step function: 0 (0.08 if STAT (n)&gt; 100

S_LPS_H (n) = f_{LSP 3 L}(STAT (n)) = Iq, 04 otherwise

The value of S_TRANS threshold used in case of transition of the LPC filters measured using S_LSP_H threshold was set at 5 0 dB.

The value of S_STAT threshold used when stationary LPC filters measured using S_LSP_L threshold was set at 2.5 dB. The threshold value G_: was set at OdB.

20 As regards the energy value characterizing a ENER_SIL silence frame, this value was set at 50 dB measured on 120 samples s (i) of the current block B i &lt;l₂0

-_cO = ENER -SIL 10.log ' <img id="imgf000044_0001" he="9" wi="20" file="imgf000044_0001.tif" img-format="tif" img-content="drawing" orientation="portrait" inline="no" />

As regards the threshold value S_FWD mentioned above and intended to further limit the tilting risk by requiring the LPC filtering mode "before"

30 when the value STAT (n) is less than this threshold, the value S_FUD was set at 60.