EP0490740A1 - Method and apparatus for pitch period determination of the speech signal in very low bitrate vocoders - Google Patents

Abstract

The method consists in splitting the speech signal, after sampling, into frames of defined duration, in carrying out a first auto-adaptive filtering (1) of the sampled signal (Sn) obtained in each frame in order to limit the influence of the first formant, in order to carry out a second filtering (2) in order to keep only a minimum of harmonics of the fundamental frequency, and in comparing (3) the signal obtained with respect to two adaptive thresholds SfMin(n) and SfMax(n), respectively positive and negative, and evolving as a function of time according to a predetermined law in order to retain only the signal portions which are respectively greater and less than the two thresholds. It next consists in calculating over a predetermined possible number of fundamental frequencies or "Pitch" M the autocorrelation of the signal obtained at the end of the preceding processing from a defined sampling instant No and in adopting for candidate values of "Pitch" M or of fundamental frequency those which are equal in number to a predetermined number n and which correspond to maxima of the autocorrelation and in applying the corresponding values from the autocorrelation into a table of scores updated on each new autocorrelation in order to retain as "Pitch" value only that which corresponds to a maximum score. <??>Application: low-bit rate vocoders. <IMAGE>

Description

The present invention relates to a method and a device for evaluating the periodicity and the voicing of the speech signal in vocoders at very low bit rate.

In known low bit rate vocoders, the speech signal is divided into frames of 20 and 30 ms so as to determine within them the periodicity still designated by "Pitch" in the English language, of the signal of speech. However during the transitions this period is not stable, and errors occur in the estimation of the "Pitch" and consequently of the voicing in these parts. Furthermore, if the speech signal is strongly noised by ambient noise, the evaluation of the "pitch" is then strongly disturbed or even erroneous.

The object of the invention is to overcome the aforementioned drawbacks.

To this end, the subject of the invention is a method for evaluating the periodicity and the voicing of the speech signal in vocoders at very low bit rate, characterized in that it consists in carrying out a first processing consisting in cutting up after sampling. the signal in frames of fixed duration, to perform a first self-adaptive filtering of the sampled signal (Sn) obtained in each frame to limit the influence of the first forming, to perform a second filtering to keep only a minimum of harmonics of the fundamental frequency, and to compare the signal obtained with respect to two adaptive thresholds SfMin (n) and SfMax (n) respectively positive and negative and evolutionary as a function of time according to a predetermined law to retain only the signal portions which are respectively higher and lower than the two thresholds and to carry out a second processing on the signal Scc (n) obtained at the end of the first processing, consi being at calculate on a predetermined number of fundamental frequencies or "Pitch" M has the autocorrelation of the signal obtained at the end of the first processing from a determined sampling instant No and to be retained for values of "Pitch" M or frequency fundamental candidates those in number equal to a predetermined number n which correspond to maxima of the autocorrelation and to record the corresponding values of the autocorrelation in a scoreboard updated with each new autocorrelation so as not to retain as value of "Pitch "than that which corresponds to a maximum score.

• la figure 1 un organigramme figurant un prétraitement du signal de parole mis en oeuvre par l'invention ;
• la figure 2 des exemples d'évolution du signal filtré et du signal final obtenu en fin de la chaîne de prétraitement de la figure 1 ;
• la figure 3 un organigramme pour le calcul de K valeurs candidates pour la détermination du "Pitch" selon l'invention ;
• la figure 4 un diagramme pour illustrer un mode de détermination du "Pitch" à partir d'un tableau de coefficients représentants différentes valeurs possibles de "Pitch" ;
• la figure 5 un diagramme illustrant le fonctionnement d'un indicateur de voisement.

Other characteristics and advantages of the invention will appear below with the aid of the description which follows given with reference to the appended drawings which represent:

• Figure 1 a flowchart showing a preprocessing of the speech signal implemented by the invention;
• FIG. 2 of the examples of evolution of the filtered signal and of the final signal obtained at the end of the preprocessing chain of FIG. 1;
• FIG. 3 a flow diagram for the calculation of K candidate values for the determination of the "pitch" according to the invention;
• FIG. 4 a diagram to illustrate a mode of determining the "Pitch" from a table of coefficients representing different possible values of "Pitch";
• FIG. 5 is a diagram illustrating the operation of a voicing indicator.

In principle, the invention consists in making, in a given frame, several "Pitch" estimates at regular intervals and in favoring successive estimates which have similar values, by giving a quality factor to each estimate. The quality factor has a maximum value when the signal is perfectly periodic and a lower value if its periodicity is less marked. As the voicing is directly linked to the autocorrelation of the speech signal for a delay equal to the value of "Pitch" retained, for a sound voiced autocorrelation is maximum while it is weak for an unvoiced sound. The voicing indication is obtained by comparing the autocorrelation with thresholds after having performed time smoothing and hysteresis operations to avoid erroneous transitions from the voiced state to the unvoiced state or vice versa.

The “Pitch” determination method comprises two main processing steps, a preprocessing step represented by the flow diagram of FIG. 1 and an autocorrelation calculation step, these two steps being able to be easily programmed on any processing processor of the known signal.

The pretreatment step is broken down as shown in FIG. 1 into a self-adaptive filtering step 1 followed by a low-pass filtering step 2 and a "self-adaptive clipping step 3".

et l'adaptation des coefficients Ai(n) est obtenue par application d'une relation de la forme : $$\text{Ai(n+1) = Ai(n) + Eps. Signe(Sb(n)xS (n-i))}$$

ou Eps est une constante de faible de valeur par exemple égale à 1/128.In the self-adaptive filtering step 1, the sampled speech signal is first whitened by a self-adaptive filter of order not too high, equal to 4 for example, so as to limit the influence of the first forming. If S (n) represents the n ^th speech sample and A _{i (n)} is the value of the i ^th coefficient the signal Sb (n) obtained at the output of the self-adaptive filter is a signal of the form: $${\text{Sb (n) = S (n) -A}}_{\text{1 (n)}}{\text{.S (n-1) -A}}_{\text{2 (n)}}{\text{.S (n-2) -A}}_{\text{3 (n)}}{\text{.S (n-3) -A}}_{\text{4 (n)}}\text{.S (n-4)}$$

and the adaptation of the coefficients Ai (n) is obtained by applying a relation of the form: $$\text{Ai (n + 1) = Ai (n) + Eps. Sign (Sb (n) xS (ni))}$$

or Eps is a low constant of value for example equal to 1/128.

The signal S _{b (n)} is then applied in step 2 to the input of a low pass filter whose role is both to keep only a minimum of fundamental harmonics and to reduce the band of signal frequency to then perform subsampling in order to reduce the execution time of the autocorrelation calculations which are described below.

ou de toute autre forme similaire capable de conférer au filtre passe bas une fréquence de coupure de l'ordre de 800 HZ, et une atténuation suffisante des fréquences au-delà de 1 000 HZ.The filtered signal Sf (n) which is thus obtained can take an equation of the form: $$\text{Sf (n) = [Sb (n) + Sb (n-9) +3 ((Sb (n-1) + Sb (n-8)) + 6 (Sb (n-2) + Sb (n-7 )) + 9 (Sb (n-3) + Sb (n-6)) + 11 (Sb (n-4) + Sb (n-5))] ./ 64}$$

or any other similar form capable of giving the low-pass filter a cut-off frequency of the order of 800 HZ, and a sufficient attenuation of the frequencies beyond 1000 HZ.

The last preprocessing which is carried out in step 3, transforms the signal Sf (n) into a signal Scc (n) by a self-adaptive clipping method of the type still known under the designation "center clipping" in the English language. -Saxon. It has the effect of reinforcing the temporal differences of the filtered signal.

In the case for example, where the signal Sf (n) contains very little fundamental at a frequency F _o and a lot of harmonic 2, the waveform which is obtained at the end of step 3 is then close to d '' a sinusoid of frequency 2.F _o presenting a slight distortion every two periods. This pretreatment of step 3 then has the effect of further reinforcing this distortion to make it easier to calculate the "Pitch" performed subsequently. This preprocessing consists, as shown in FIGS. 2A and 2B, of calculating two adaptive thresholds, SfMin (n) and SfMax (n), which change over time, so as to retain only the signal portions which are respectively lower and greater than these two thresholds.

$$\text{SfMax(n) = E.SfMax(n-1)}$$

$$\text{avec E = exp (-Te/Tau)}$$

où Te est la période d'échantillonnage et Tau est une constante de temps de l'ordre de 5 à 10 ms.The thresholds SfMin (n) and SfMax (n) check the relationships: $$\text{SfMin (n) = E.SfMin (n-1)}$$

$$\text{SfMax (n) = E.SfMax (n-1)}$$

$$\text{with E = exp (-Te / Tau)}$$

where Te is the sampling period and Tau is a time constant of the order of 5 to 10 ms.

Si Sf(n)>SfMax(n) alors la différence Sf(n)-SfMax(n) est amplifiée pour donner un signal Scc(n) défini suivant la relation : $$\text{Scc(n)=G[Sf(n)-SfMax(n)]}$$

Dans ce cas l'ancienne valeur de SfMax(n) est actualisée par la nouvelle valeur de Sf(n) et SfMax(n) est rendue égal à Sf(n). Par contre si Sf(n)<SfMin(n) c'est la différence Sf(n)-SfMin(n) qui est amplifiée pour donner un signal
Scc(n) défini suivant la relation : $$\text{Scc(n)=G[Sf(n)-SfMin(n)]}$$

et l'ancienne valeur de SfMin(n)=Sf(n) est actualisée par la nouvelle valeur de Sf(n).It follows from the above that the signal Scc (n) obtained at the end of the execution of step 3 is always of zero amplitude except for: $$\text{SfMax (n) <Sf (n) <SfMin (n)}$$

If Sf (n)> SfMax (n) then the difference Sf (n) -SfMax (n) is amplified to give a signal Scc (n) defined according to the relation: $$\text{Scc (n) = G [Sf (n) -SfMax (n)]}$$

In this case the old value of SfMax (n) is updated with the new value of Sf (n) and SfMax (n) is made equal to Sf (n). On the other hand if Sf (n) <SfMin (n) it is the difference Sf (n) -SfMin (n) which is amplified to give a signal
Scc (n) defined according to the relation: $$\text{Scc (n) = G [Sf (n) -SfMin (n)]}$$

and the old value of SfMin (n) = Sf (n) is updated by the new value of Sf (n).

In equations (7) and (8) G represents a gain value which is preferably chosen constant to improve the calculation accuracy for the case where a signal processor working in fixed point is used.

For the case where in the previous relationships the value of the time constant Tau is chosen to be zero, it goes without saying that the signal Scc (n) is identical to the signal Sf (n).

The following autocorrelation calculation step is carried out for each value M of the "Pitch" for a determined position Sampling no. In the description which follows, the calculation takes place by means of a sub-sampling of a factor of 4 over a time range of 160 samples corresponding to a maximum value which can be accepted for the "pitch". It is obvious that the same principle can still apply for a different sampling order and for another different range.

$$\text{RMM=Scc(No-M)**2+Scc(No+4-M)**2+Scc(No+8-M)+...+Scc(No+-160-M)**2}$$

$$\text{ROM=Scc(No).Scc(No-M)+Scc(No+4).Scc(No+4-M)+...+Scc(No+1-60).Scc(No+160-M)}$$

The calculation consists, as represented by steps 4 to 6 on the flow diagram of FIG. 3, of calculating three quantities Roo, RMM and ROM defined as follows, in which the sign ** denotes an increase in power. $$\text{Roo = Scc (No) ** 2 + Scc (No + 4) ** 2 + Scc (No + 8) ** 2 + ... + Scc (No + 160) ** 2}$$

$$\text{RMM = Scc (No-M) ** 2 + Scc (No + 4-M) ** 2 + Scc (No + 8-M) + ... + Scc (No + -160-M) ** 2}$$

$$\text{ROM = Scc (No) .Scc (No-M) + Scc (No + 4) .Scc (No + 4-M) + ... + Scc (No + 1-60) .Scc (No + 160-M )}$$

For each position No, chosen the quantity Roo is calculated in step 4 only once, the quantity RMM is calculated entirely in step 5 only for certain values of M and by iteration for the others, and the ROM quantity is calculated, in full in step 5 for each value of M.

The values of M for which the autocorrelation calculation takes place correspond to a fundamental frequency of the speech signal which can vary between 50 HZ and 400 HZ. These are delimited on three ranges defined as follows:
Range 1 M = 20, 21, 22 ...... 40 i.e. 21 values in step 1
Range 2 M = 42, 44, 46 ...... 80 i.e. 20 values in step 2
Range 3 M = 84, 88, 92 ...... 160 i.e. 20 values in step 4
that is to say a total of 61 different values which can be coded for example on 6 bits with a minimum precision of 5% corresponding to the value of a semitone of the chromatic range.

The iteration formula used for the RMM calculation is as follows: $$\text{RMM (M) = RMM (M-4) + Scc (No-M) ** 2-Scc (No + 164-M) ** 2}$$

Furthermore, to improve the accuracy of finding the autocorrelation maxima, a parabolic interpolation formula is used which, for a given value M, uses the values of the preceding quantities for M-dM, M, and M + dM, dM being a step value equal to 1, 2 or 4 depending on the range considered. It follows that only the values of RMM (19), RMM (20), RMM (21), and RMM (22) are to be fully calculated, the others are iterated, including for M = 164.

et Rau(M) = ROM(M)**2/[ROO(M).RMM(M)] si ROM(M)> 0Depending on the above, we calculate a value: Rau (M) defined as follows: $$\text{Rau (M) = 0 if ROM (M) <= 0}$$

and Rau (M) = ROM (M) ** 2 / [ROO (M) .RMM (M)] if ROM (M)> 0

sont prises en considération à l'étape 6. Pour ces seules valeurs de M il est calculé ensuite une valeur Rint interpolée paraboliquement, suivant la relation : $$\text{Rint = Rau(M) + 1/8 [Rau(M+dM) - Rau(MdM)]**2 / [2.Rau(M) - Rau(M-dM) - Rau(M+dM)]}$$

pour ne retenir dans la suite des traitements que les K valeurs correspondant aux K valeurs les plus élevées de Rint (et les valeurs associées de M), par exemple les K=2 maxima les plus importants, notés Rmax (1) , ..., Rmax (K) (et Mmax (1), ..., Mmax (K)).Only the values of M for which a local maximum is obtained, namely those for which Rau (M) checks the inequalities: $$\text{Rau (M)> Rau (M-dM) and Rau (M)> = Rau (M + dM)}$$

are taken into consideration in step 6. For these values of M only, a Rint value is parabolically interpolated, according to the relation: $$\text{Rint = Rau (M) + 1/8 [Rau (M + dM) - Rau (MdM)] ** 2 / [2.Rau (M) - Rau (M-dM) - Rau (M + dM)]}$$

to retain in the following processing only the K values corresponding to the K highest values of Rint (and the associated values of M), for example the most important K = 2 maxima, denoted Rmax (1), ... , Rmax (K) (and Mmax (1), ..., Mmax (K)).

The continuation of the processing consists in keeping up to date a scoreboard associated with the different possible values for the "Pitch" M.

This table, denoted Score (i), in FIG. 4 contains for the i = 1 to 61 M values of "Pitch" an amount which is an increasing function of the degree of likelihood of the associated "Pitch" (from 20 to 160), and which is updated with each new evaluation of the autocorrelations (typically every 5 to 10 ms), taking into account the fact that, from one evaluation to the next, the positions of the maxima can vary by more than one unit, remain stationary or vary by at least one unit depending on whether the "Pitch" is respectively increasing, stationary, or decreasing.

$$\text{ExScore (i) = Score (i) pour i = 2}$$

$$\text{et ExScore (62) = 0}$$

The scoreboard is transferred to a temporary table, ExScore (i) note not shown. This table is defined according to the values of i as follows: $$\text{ExScore (0) = 0}$$

$$\text{ExScore (i) = Score (i) for i = 2}$$

$$\text{and ExScore (62) = 0}$$

avec $$\text{ScoreMin = MIN [Score (20)), Score (21), ..., Score (61)]}$$

Periodically (if not systematically), the minimum value is withdrawn to avoid possible overflows in such a way that: $$\text{ExScore (i) = ExScore (i) - ScoreMin}$$

with $$\text{ScoreMin = MIN [Score (20)), Score (21), ..., Score (61)]}$$

The different scores are initialized to take account of a possible "Pitch" drift, which gives: $$\text{Score (i) = MAX [ExScore (i-1), ExScore (i), ExScore (i + 1)] for i = 20, ..., 61}$$

$$\text{pour k = 1, 2, ..., K.}$$

$$\text{et i = I(1), ..., I(K)}$$

Finally, for the values I (1), ..., I (K) of i corresponding to the K Pitchs Mmax (1) ... Mmax (K) where maximums are encountered, the scores are increased by an equal amount at the maxima of the autocorrelation found such that $$\text{Score (I (K)) = Score (I (K)) + Rmax (K)}$$

$$\text{for k = 1, 2, ..., K.}$$

$$\text{and i = I (1), ..., I (K)}$$

Finally, the Pitch value M retained for the NO position is that corresponding to the maximum of the scoreboard, ScoreMax, located at the Imax index in this table.

If, for reasons of calculation accuracy and / or algorithm, several successive values of the score are equal to the maximum ScoreMax, namely Score (Imax), Score (Imax + 1), Score (Imax + dI), the value retained for the Pitch is that which corresponds to Imax + [dI / 2], [dI / 2] being the integer value of the division dI by 2, as indicated in figure 4.

For a given frame, or the calculations described above are carried out several times, the final value of the Pitch is that obtained at the last iteration, it being understood that there are between 2 and 4 iterations per frame.

The Pitch value M which is thus obtained corresponds to the most likely periodicity of the speech signal centered around the position N _o with a resolution of 1, 2 or 4 depending on the range where the value of M. is located. voicing is then calculated by performing an autocorrelation, normalized for a delay equal to M and possibly for neighboring values if the resolution is greater than 1, of the original speech signal S (n) and not on the pretreated signal Scc (n) as for the Pitch calculation.

For example, for M = 30, the normalized autocorrelation is calculated only for a delay of 30, for M = 40 , it is calculated for delays of 40 and 41 and for M = 100 it is calculated for delays of 100, but also for delays of 98, 99 as well as 101 and 102 (the resolution being 4 for M = 100).

ou $$\text{R = 0 si ROM est plus petit ou égal à zéro}$$

$${\text{Roo = S(N}}_{\text{o}}{\text{)**2+S(N}}_{\text{o}}{\text{+1)**2+...+S(N}}_{\text{o}}\text{+160)**2}$$

$${\text{RMM = S (N}}_{\text{o}}{\text{-M)**2+S (N}}_{\text{o}}{\text{+1-M)**2+...+S(N}}_{\text{o}}\text{+160-M)**2}$$

$${\text{ROM = S(N}}_{\text{o}}{\text{).S(N}}_{\text{o}}{\text{-M)+S(N}}_{\text{o}}{\text{+1).S(N}}_{\text{o}}\text{+1-M)+...}$$

$${\text{+S (N}}_{\text{o}}{\text{+160).S(N}}_{\text{o}}\text{+160-M)}$$

In all cases, the value retained Rm is the largest of the values thus calculated, an elementary value for M given being defined by the relations: $$\text{R = <figure-callout id="2" label="ROM" filenames="imgf0002.png,imgf0003.png" state="{{state}}">ROM</figure-callout> 2 / (Roo.RMM) if ROM is positive}$$

or $$\text{R = 0 if ROM is less than or equal to zero}$$

$${\text{Roo = S (N}}_{\text{o}}{\text{) ** 2 + S (N}}_{\text{o}}{\text{+1) ** 2 + ... + S (N}}_{\text{o}}\text{+160) ** 2}$$

$${\text{RMM = S (N}}_{\text{o}}{\text{-M) ** 2 + S (N}}_{\text{o}}{\text{+ 1-M) ** 2 + ... + S (N}}_{\text{o}}\text{+ 160-M) ** 2}$$

$${\text{ROM = S (N}}_{\text{o}}{\text{) .S (N}}_{\text{o}}{\text{-M) + S (N}}_{\text{o}}{\text{+1) .S (N}}_{\text{o}}\text{+ 1-M) + ...}$$

$${\text{+ S (N}}_{\text{o}}{\text{+160) .S (N}}_{\text{o}}\text{+ 160-M)}$$

Unlike the previous calculation method implemented to calculate the signal S _cc (n), the signal S (n) is not undersampled.

The quantity Roo does not depend on M and is calculated only once. It is possible to be satisfied with calculating RMM for the only nominal value of M, namely that provided by the Pitch calculation method described above and for values close to M to calculate RMM by iteration if necessary. However, the ROM quantity must be calculated for each of the values of M.

où a est une constante égale de préférence à ¼ ou ½ pour que les performances soient satisfaisantes.To limit fluctuations, especially in a noisy environment, the maximum value of the quantity R _m thus obtained is filtered by a low-pass filter between two successive passages (corresponding to two successive values of the reference value N _o ), for obtain a filtered value Rf (P) defined at each iteration p by the relation: $${\text{Rf (P) = (1-a) .Rf (P-1) + aR}}_{\text{m}}$$

where a is a constant preferably equal to ¼ or ½ for the performances to be satisfactory.

By tolerating a delay in coding an even more satisfactory expression may still be the following $${\text{Rf (P) = [R}}_{\text{m}}{\text{(P-1) + 2R}}_{\text{m}}{\text{(P) + R}}_{\text{m}}\text{(P + 1)] / 4}$$

Finally, the quantity Rf (P) is compared as shown in FIG. 5 to two thresholds S _V and S _NV called respectively the voicing and non-voicing threshold such that the threshold S _V is greater than the threshold S _NV to obtain a binary indicator voicing IV as shown in Figure 5.

In Figure 5,

state IV = 1 corresponds to a voiced sound and state IV = 0 corresponds to an unvoiced sound.

Starting from state IV = 1, IV goes to state 0 when Rf (P) becomes lower than SNV and starting from state IV = 0, IV goes to state 1 when Rf (P) becomes greater than SV.

Typical values for adjusting the two thresholds SNV and SV can for example be fixed at SV = 0.2 and SNV = 0.05 by taking 1 for maximum value of Rf (P) and 0 for minimum value of Rf (P).

In order to optimize the performance of the voicing decision, it is preferable that these thresholds be adjustable to give a certain inertia to the decision which is not perceptible at hearing to avoid local errors in the appreciation of the voicing.

Claims (6)

Method for the evaluation of the periodicity and of the voicing of the speech signal in vocoders at very low bit rate characterized in that it consists in carrying out a first processing consisting in cutting up after sampling the signal into frames of determined duration, in carrying out a first self-adaptive filtering (1) of the sampled signal (Sn) obtained in each frame to limit the influence of the first component, to perform a second filtering (2) to keep only a minimum of harmonics of the fundamental frequency, and to compare (3) the signal obtained even with respect to two adaptive thresholds SfMin (n) and SfMax (n) respectively positive and negative and evolutionary as a function of time according to a predetermined law to retain only the signal portions which are respectively greater and lower than the two thresholds and to carry out a second processing (4, 5, 6) on the signal S _c c (n) obtained at the end of the first processing, consisting in calculating r on a predetermined number of fundamental frequencies or "Pitch" M possible the autocorrelation of the signal obtained at the end of the first processing from a determined sampling instant No and to be retained for values of "Pitch" M or frequency fundamental candidates those in number equal to a predetermined number n which correspond to maxima of autocorrelation and to record the corresponding values of autocorrelation in a scoreboard updated with each new autocorrelation so as not to retain the value of "Pitch""than that which corresponds to a maximum score. Method according to Claim 1, characterized in that the autocorrelation of the signal Scc (n) obtained at the end of the first processing is calculated from the sampling instant No on a determined number of samples which follows it by carrying out: - a first addition (ROO) of a first series of samples separated from each other by a determined number of samples; - a second addition (RMM) of a second series of samples each corresponding to a sample of the first series shifted late by the value of "Pitch"M; - a third addition (ROM) of respectively sample products from the first suite with their counterpart in the second suite so as to perform the quotient (RauM) of the result (ROM) of the third addition by the product of the other two (ROO x RMM) to consider only a determined number K of values of M for which the quotient Rau (M) is maximum locally. Method according to Claims 1 and 2, characterized in that it consists in evaluating the voicing by calculating the autocorrelation of the sampled speech signal for a delay equal to the value of the "Pitch" M retained and the neighboring values so as to retain only the larger of the values thus calculated, to perform a low pass filtering of this value and compare it with hysteresis to two thresholds of voicing and non voicing respectively to decide on the voiced or unvoiced state of the speech signal. Method according to any one of Claims 1 to 3, characterized in that the first self-adaptive filtering consists in subtracting from each current sample Sn the sum weighted by coefficients Ai (n) from a determined number i of previous samples, the adaptation of the coefficients Ai (n + 1) being obtained by adding to the current coefficient Ai (n) a quantity EPS affected by a sign equal to the signal of the result of the subtraction by the sign of the sample S (ni). Method according to Claim 4, characterized in that the two adaptive thresholds SfMin (n) and SfMax (n) are determined for each current sample at time n from the previous sample of time n-1 by the relationships : $$\text{SfMin (n) = E. SfMin (n-1)}$$

$$\text{SfMax (n) = E. SfMax (n-1)}$$

where E is an exponential function of the ratio between the period Te of the samples and a constant Tau between 5 and 10 ms. Device for the evaluation of the periodicity and of the voicing of the speech signal in vocoders at very low bit rate characterized in that it comprises a signal processing processor programmed according to any one of claims 1 to 5.

Images