FR2545301A1 - Method of compressing/decompressing a signal of vocal type - Google Patents

Abstract

Method of compressing/decompressing a signal of digitised vocal type ensuring, through a first transcoding, a reduction in the amount of data required in the subsequent reconstitution of the signal and then subsequently, through a second transcoding, the reconstitution of the said signal from the compressed data. The amplitude samples corresponding to the extrema of the initial signal are transcoded into binary extremum words each containing, on the one hand, an indication relating to the amplitude of an extremum, and on the other hand an indication relating to the time lapse separating the relevant extremum from the preceding one, the succession of samples corresponding to a state of temporary silence of the initial signal is transcoded, following recognition of this state, by a possibly refreshed item of binary information which includes a descriptor and an indication of duration. The extremum words and the silence information enable the signal to be reconstituted subsequently in the form of amplitude samples. The invention applies in particular to the storing of PCM coded signals and to the transmitting of such signals in packets.

Description

Voice decompression compression method
The present invention relates to a compression-decompression method of a voice type signal and in particular of a voice type signal sampled and digitized, for example according to the so-called pulse modulation and coding technique.

 This method is intended to facilitate the temporary or permanent storage of the signal and possibly its transmission by ensuring a limitation of the amount of data necessary for its reconstruction after storage and / or transmission.

 Indeed, it is known that the storage in digital form of a voice type signal conventionally leads to taking into account a large number of data which are not all essential for the subsequent reproduction of the initial signal, in particular when the we are looking for a readable restitution of the signal rather than an identical restitution.

 Thus for example the technique known as pulse modulation and MIC coding makes it possible to code a voice type signal by sampling at a frequency of eight kilohertz and coding of each sample according to its amplitude in the form of a byte. Such a technique which allows satisfactory restoration of voice type signals, on the other hand has the drawback of requiring large volumes of memory when it is desired to keep a signal of relatively long duration.

 This therefore leads to the search for means making it possible to reduce the amount of data necessary for the reproduction of the initial signals while maintaining a suitable quality of reproduction.

 In general, we already know from the French patent 22 52 799 of Mrs Bertrand de COSNAC and Xavier RODET a process allowing in the first place to record a digitized signal by a series of pairs of information in which each pair of information includes an indication relating to the amplitude of an extremum of the signal and an indication relating to the lapse of time separating this extremum from the preceding extremum then secondly to reconstitute at least the essential characteristics of the initial signal.

 The present invention therefore provides a compression-decompression method applying the teaching of the patent mentioned above to the processing of a voice type signal in order to compress in a first phase a voice type signal, coded in the form of a sequence of amplitude samples according to a technique known as pulse modulation and coding, and in a second decompression phase to subsequently restore at least the significant constituents of the initial signal.

According to a characteristic of the invention, the amplitude samples corresponding to the extrema of the coded signal are transcoded into binary extremum words each comprising on the one hand an indication relating to the amplitude of an extremum, obtained from of the corresponding amplitude sample, on the other hand an indication relating to the period of time separating this sample from the sample corresponding to the preceding extremum and in that the succession of samples corresponding to a state of silence is transcoded of the coded signal, following recognition of this state, by binary silence information composed of at least one word and possibly repeated, which includes a descriptor and an indication relating to the duration of the silence, so as to reduce the volume of binary data necessary for the subsequent reconstruction of the signal
According to another characteristic of the invention, the generation of silence information is triggered by the overflow of a duration counter incremented at each sample acquisition in the established silence phase and in that the generation of information at the end of silence is triggered by exceeding a silence output threshold of a first bounded up-down counter incremented with a first increment step by samples whose amplitude is greater than an output setpoint of silence and decremented with a first decrementing step less than the first incrementing step by the samples whose amplitude is less than the silence output setpoint value.

 According to another characteristic of the invention, the entry into the established silence phase defined by a word of zero amplitude is triggered by exceeding a silence entry threshold of a second up-down counter incremented with a second step incrementation by samples whose amplitude is less than a silence input setpoint and decremented with a second incrementation step greater than the second incrementation step by samples whose amplitude is greater than the value of silence entry instruction.

According to another characteristic, the indications of amplitude of extremum words obtained from the corresponding amplitude samples are coded in one or the other of at least two homothetic ranges individually selected as a function of the increasing energy. of the coded signal which is measured from the amplitude samples
According to another characteristic during the reconstitution of the initial signal coded after the transcoding phase previously claimed, the amplitude samples sought are obtained by a reverse transcoding phase of the indications of amplitude of the extremum words and by a reverse coding after linear interpolation between extrema for the amplitude samples to be inserted between extremum words, said linear interpolation being carried out from two pairs of calculation values obtained from the amplitude indications in linear form and from the time lapse determined from the relative indications to two successive extremum words for the sample preceding the one being calculated and for the second in time of the two extremum words considered.

 The invention, its characteristics and its advantages are specified in the following description in relation to the figures listed below.

 Figure 1 shows a block diagram relating to the implementation of the method according to the invention.

 FIG. 2 presents a diagram of a compression-decompression equipment for implementing the method according to the invention.

 Figures 3 and 4 each present a diagram relating to the detection and treatment of silences.

 FIG. 5 shows a diagram relating to a range selection processing phase.

 FIG. 6 presents a diagram relating to a phase of coding of exztrema.

 FIG. 7 presents a diagram relating to a phase for processing the silence information during reverse transcoding.

 FIG. 8 presents a diagram relating to a phase of reading the direct transcoded signal.

 Figure 9 shows a diagram relating to a sample calculation phase.

 FIG. 10 presents a diagram relating to a phase of coding MIC of sample.

 The method according to the invention relates to low frequency signals and in particular voice type signals incorporating significant constituents interspersed with rests, it involves coding of these signals in the form of a series of amplitude samples by the implementation. gives technique called pulse modulation and coding.

 By way of example in the following description, the samples of a coded signal MIC which are conventionally supplied every one hundred and twenty five microseconds by a time channel in the form of a byte are taken into account. In a known manner, this byte corresponds to a sample of the signal coded according to law A as defined by Opinion G711 of CEINT.

 According to the invention, the MIC coded signal supplied by an incoming MIC channel - FIG. 1 - undergoes a so-called direct transcoding phase which is symbolized by the TCD block and which allows compression of data in volume.

 This phase of direct transcoding essentially ensures a conversion of the succession of amplitude sample of a signal coded there in MIC into a reduced succession of words or binary information based on the extremes of the coded signal This conversion is carried out in agreement with the process which is the subject of French patent 22 59 799 already cited, as a function of the energy of the coded signal and taking account of the silences included in this signal, in particular when it is a speech signal.

For this purpose, the direct transcoding block TCD comprises three modules
MESE, CDEX, DTSI which from the incoming MIC encoded signal via an incoming MIC channel respectively provide the following results - a measurement of the energy of the encoded signal as defined by the succession of MIC samples from the incoming MIC channel up to to the MESE energy measurement module, - a detection and a processing of the silences contained in the coded signal as defined by the same samples coming up to the DTSI silence detection module, - a transcoding of the coded signal MIC taking into account the same MIC samples received and indications of silence detection and energy measurement at the CODEX coding module
The indications provided by the three MESE, DTSI, CDEX modules are transmitted to a temporary storage and multiplexing unit EMB intended to constitute a direct transcoded signal constituting the replica of the initial coded signal Ie for the purposes of temporary storage and / or transmission.
The indication TRAINS symbolizes the temporary or possibly final conservation phase of the direct transcoded signal, before transmission for the reverse transcoding phase symbolized by the TCI block.

 The reverse transcoding phase makes it possible to reconstruct a coded MIC signal from a direct transcoded signal obtained in the manner mentioned above.

 For this purpose, the reverse transcoding block TCI comprises a demultiplexing and temporary storage unit UDB making it possible to separate the indications coming from the MESE, DTSI and CDEX modules from the direct transcoding block TCD.

The TCI reverse transcoding block also includes three RSTE, DCSL and DCSI modules which correspond respectively to the MESE, CDEX and DTSI modules and which therefore respectively provide the PeSt Itution of energy hiding indications for the RSTE module, the deoding of the silence indications for the DTSI module and decoding of the direct transcoded signal for the DCSL module in conjunction with the RSTE energy restitution and DTSI silence decoding modules, so as to reconstitute at least the essential characteristics of the signal in MIC form initial coded
The MIC samples thus produced are then liable to be retransmitted on a channel of a usual outgoing MIC link.

 In practical terms, the intermediate transcoding method according to the invention can be implemented using specialized transcoding equipment which can be simple and inexpensive, such as that organized around a microprocessor shown in FIG. 2.

The transcoding equipment presented here is connected to an incoming link DR which supplies it with the samples MIC of the coded signal at the rate of the signals of an external clock HD and with a frame synchronization SY. These HD external clock and frame synchronization signals are also used for retransmission on an outgoing link of the initial signal reconstituted in the form of samples.
MIC after complete intermediate transcoding.

 The binary elements of the samples transmitted by the incoming link DR are received one by one by a serial-parallel input register RE which makes them available to the microprocessor MP, sample by sample, in a buffer register TE addressed in reading by this microprocessor MP. In the example chosen, the microprocessor is of the eight bit type and it is capable of receiving in parallel the byte that constitutes a sample in one of its non-represented input-output registers.

 An identical organization makes it possible to restore samples MIC or possibly bytes by means of a buffer register TS connected at the output of the microprocessor MP via the input-output links of the latter as well as the buffer register TE.

 A parallel-serial output register RS connected to the parallel outputs of the buffer register TS ensures the transmission of the bytes received from the microprocessor, when these bytes are intended to be retransmitted on an outgoing link DT, of the MIC type, as shown in FIG. 2. output register BS is conventionally controlled by the clock clock HD and synchronization signals SY applied to the input register BE.

 As a variant, it is conceivable to transmit outwards for storage and / or transmission and by means of the register RE, the words or binary information obtained at the end of the direct transcoding phase by the microprocessor. The exit control by the output register RS is then ensured by the microprocessor MP itself, for example from the elaboration of a word or of binary information which are then preferably equal or multiple of a byte.

The mieroproeesseur MP synchronized by a local clock HI is conventionally associated by the transcoding operations with a read-only memory of program PR conventionally of the PROM type and with a random access memory RA conventionally of the RAM type. This lIA memory or an extension of the latter can possibly serve as temporary memory for the words or binary information obtained at the end of the direct transcoding phase, this normally does not correspond to the variant mentioned above.
The memories are conventionally addressed by the microprocessor MP via an address bus AD, are read and written under the control of a command link CT and transmit their content via a data bus DI. A buffer register TT allows a possible exchange information between the microprocessor PS and a standard operating terminal.

 In direct transcoding phase, the A-coded MIC samples which are received at the rate of the external HD clock every 125 microseconds, are processed according to the process defined below, which comprises three processing phases SCS, SCP, SCE corresponding respectively to the DTSI silence detection and processing modules, MESE energy measurement and CODEX coding modules, defined above.

 The processing of the silences of the original signal coded MIC by the DTSI module is schematically presented in FIGS. 3 and 4 FIG. 3 relates to the processing of the phases of silence established, that is to say during startup and during silence, FIG. 4 relates entering the silence phase during the signal after issuing extremum words.

 The processing of the phases of silence established, referenced SCS1 and presented in FIG. 3, starts with an operation of acquiring a sample requiring beforehand an operation of searching for a SY synchronization reference mark MIC transmitted by the incoming link such as DR in the figure. 1, so as to select the time channel whose samples are to be transcoded using the method according to the invention.

 A sample MIC being assumed to be acquired, its amplitude is determined by a comparison operation with respect to an output setpoint value Af which is chosen as a function of the current energy range.

 In a preferred embodiment comprising three homothetic energy ranges, there are therefore three output setpoint values Af and at a given instant only one of the three ranges is considered, this is what is called the current range. In practice, only the amplitudes of the A-coded MIC samples are considered and the sign of these samples is neglected, which therefore amounts to having only one output setpoint threshold Af per range instead of two.

If the amplitude of an acquired sample is lower than the current output threshold, a limited up-down counter known as the silence output is decremented insofar as this is not already at the countdown limit, i.e. - say zero, and it being understood that any emission of samples conventionally starts with a silence phase. If the amplitude of a sample is greater than the output setpoint value Af, the silence output decoder-detector is incremented with an incremental step preferably greater than the corresponding decrementing step, so as to favor the useful signals over the silenees and not to take into account short-term parasites;
If the increment of the silence coepteuP-de-counter does not cause an overflow of this counter, a second counter, called the duration of silence, which is also incremented when there has been in acquisition in silence phase of an amplitude sample, is incremented lower than the current setpoint.

 If the incrementation of the silence duration counter does not cause this counter to overflow beyond a duration counting threshold Cd the operation continues by returning to the operation of search for loss of synchronization prior to the acquisition of a sample.

 If the incrementation of the silence duration counter causes an overflow beyond the duration counting threshold Cd, the microprocessor generates silence information 151 which in a preferred embodiment comprises two one-byte words together forming a descriptor composed of six bits and an indication relating to the duration of the silence composed of ten bits is therefore at most one hundred and twenty eight milliseconds.

 Following the production of such silence information IS1, the silence duration counter is reset to zero and the operation continues by returning to the signaling search operation prior to the acquisition of a sample.

 If, following the incrementation of the silence counter, the silence output threshold Ss is reached, an end of silence information IS2 is generated, it is presented in a form analogous to the silence information IS1 and therefore comprises a set of two bytes comprising the same descriptor as for the silence information IS1 and an indication of silence duration corresponding to the content of the silence duration counter. The different different silence counters are then reset to zero.

 The processing entered in the silence phase referenced SCS2 (FIG. 4) can only be conceived after a phase of coding MIC samples of coded signal having had amplitudes greater than the nettle setpoint threshold Af, that is to say after a phase of processing by the CDEX coding module.

 The processing for entering the silence phase SCS2 starts with a sample acquisition operation involving an operation for searching for a synchronization reference point MIC transmitted by the incoming link DB, of a wheel that the processing of the phases of silence established.

 Following this sample acquisition, the amplitude of the latter is compared with an entry setpoint value in the silence phase Ad, just as previously, there is therefore an entry setpoint value Ad for each range and the value taken into account is that of the current range.

 If the amplitude of the current MIC sample is not less than the input setpoint, a bounded down counter of silence input is decremented by one step, insofar as this counter is not already at the countdown limit; the sample is then subjected to the SCE coding phase.

 If the amplitude of the current MIC sample is less than the input setpoint value Ad, the silence input up-down counter is incremented with an increment step less than the deereration step to again favor the signals useful in relation to silence.

 If the silence input count does not exceed a silence input threshold value Es, the sample MIC is subjected to the coding phase SCE, otherwise this phase is abandoned and there is generation of a extremum word, in this case a byte, comprising an indication of zero amplitude and a time indication corresponding to the period of time separating this sample from the sample having led to an extremum or to a previous pseudo-extremum, this term being defined later.

 In addition, it is checked that the current range is identical to the previous range, according to a process also discussed below so as to possibly generate a range change word. After resetting the silence counters, the subsequent MIC samples will be at least temporarily no longer subjected to this entry processing in the SCS2 silence phase, but to the processing of the established SCS1 silence phases.

 The energy measurement of the signal from the MIC samples by the MESSE module essentially aims to determine that of the coding ranges to be implemented for the coding of the MIC samples by the CDEX coding module, it is schematically presented in FIG. 5.

 This measurement is included in a range selection process SCP which involves a sample acquisition operation MIC preceded by a synchronization mark search operation MC, it makes it possible to determine the current coding range of the samples in words of ex- extremum.

 For this purpose, each MIC sample is firstly rectified by masking its sign bit, and secondly filtered by means of a low-pass filter which provides the average value of the energy of the coded signal.

Preferably, a filter with a time constant large enough to give a long-term value of the energy is produced. Preferably also the value of the energy must vary fast enough at the end of a silence to avoid excessive correction of the signal.

The filter is therefore adapted to the amplitude of the signal and has a variable input gain, the adaptation criterion is based on a comparison of the signal and the energy measured so that the gain of the filter is unitary when the signal amplitude is less than the measured energy and the gain of the filter is greater than unity when the amplitude of this signal is greater than the measured energy
In a preferred embodiment this low-pass filter is a first-order digital filter conventionally defined by an equation of the form Y (z) = GX (z) + B1 ç Y (s). z 1 1 in which G is the input gain and B1 a weighting coefficient of cosine shape as a function of the filter cutoff frequency and the sampling frequency.

 The response time of the filter is checked by varying its gain after comparing the amplitude of the current sample with the previous one.

 The energy level reached by the samples makes it possible to determine the range of coding of the samples in extremum words.

 In order to obtain a precise definition of the word amplitudes dsextremumS for both low and high amplitudes, several coding ranges are provided with different scales and respectively equivalent input and output silence threshold values; as seen in the embodiment shown, three ranges are used.

 The coding phase of samples 11IC with words of extremum referenced SCE carried out by the CDEX module is presented in FIG. 6.

 Firstly, it includes a change in conversion law allowing the amplitude of a MIC sample coded in logarithmic law to be translated into increasing binary form.

 A second operation takes into account the direction of variation of the amplitudes of successive samples in order to determine whether the sample varies in the direction determined for the previous sample so as to detect the appearance of an extremum of the coded signal. which results in a change of direction of variation for the current sample and which indicates that the previous sample corresponds to a signal extremum.

In the event of a change of direction of variation, the amplitude of this variation is compared with a setpoint value of current amplitude
Sa which, like the silence input and output setpoint values -Ad and Af, depends on the current coding range.

 If the amplitude variation is greater than the current amplitude setpoint, the generation of an extremum word is initiated.

 For this purpose, the amplitude of the sample corresponding to the extremum is transcoded via the current coding table to an indication of extremum amplitude intended to be integrated into an extremum word.

 In the example described here, this extremum word referenced M1 is in the form of a byte, five bits of which are assigned to the amplitude indication and the other three bits of which are assigned to the indication of the period of time. separating the current extremum sample, the amplitude indication of which is attached, from the previous extremum sample.

 This period of time is counted by an extremum word duration counter reset to zero with each generation of extremum word, or pseudo extremum as indicated below.

 If the amplitude variation of the current sample compared to the previous sample is less than the current amplitude setpoint, while there has been a change in direction of variation, the word duration counter is incremented extremum; as long as the counter has not reached a counting threshold determined below, an SCP range selection processing phase is again undertaken.

 If the extremum word duration counter overflows beyond a counting threshold Sm corresponding to the three time lapse indication bits, the generation of an extremum word M1 is initiated.

 If, on the other hand, the variation in the amplitude of the current sample is within the variation determined for the previous sample, the extremum word duration counter is incremented. If in the latter case the counter overflows beyond the counting threshold Sm, the generation of a pseudo-extremum word is initiated, otherwise the processing phase of choice of range SCP is again undertaken after acquisition of the sample Next HIC.

 The pseudo extremum word is of the same structure as the extremum word M1 and therefore includes an indication of the amplitude of the MIC sample preceding the current sample which is coded on five bits and the indication of the maximum time lapse likely to be indicated by the three bits.

 The generation of an extremum word or pseudo extremum causes the reset of the extremum duration counter and a verification of range choice comparing the current range determined by the current sample with the previous range determined for the sample of extremum or pseudo extremum which has just been produced. If the energy measurement carried out in these two cases led to two different ranges, a range change word M2 is produced. This word M2 essentially comprises a descriptor composed of five bits and an indication of range change on three bits.

 In the case mentioned above where there was generation of a range change word M2 and in the case where the last energy measurement led to the choice of the sema range only for 19 extremum or the previous pseudo extremum, a SCP range selection processing is again undertaken for the next MIC sample.

 At the end of the coded signal, an end word is generated in the form of a specific byte Fi.

 It is therefore possible to code an initial signal coded MIC in a succession of words d t extreugy of psaudo-extremum, of change of range and of end, here constituted by oets9 and silence information, here each constituted by two bytes. Such a succession can be stored temporarily or permanently for the purpose of a subsequent reconstruction of the initial signal.

This allows, for example, a recording of massages received by an automatic telephone answering machine or by an information center and their restitution, possibly at accelerated speed, using a static mescire of significantly lower capacity than that required for the same recording. from messages received in coded form
MIC.

 Such direct transcoding is also well suited to the transmission of information in the form of packets, insofar as it ensures a high concentration of the data to be transmitted.

 The reverse TCI transcoding making it possible to restore MIC samples corresponding to the initial signal essentially comprises four processing phases using the three modules for restoring the RSTE energy indications, decoding the DCSI silence indications and decoding the direct transcoded signal. DCSI, the latter mainly dealing with the words extrema and pseudo-extresa.

 A first phase SDS of reverse transcoding relates to the processing of silence information, it is divided into two sub-phases of processing SDS1 and SDS2 presented in FIG. 7.

 The SDS1 sub-phase corresponds to the processing of the words of zero amplitude generated in the phase of silence at the start of the initial signal, to the presentation of the first zero sample t it is firstly determined whether the restitution should be carried out at the same speed as l 'registration or if a prompt return is requested.

 In the event of a request for rapid restitution, a search for the last silence information of the initial phase of silence is undertaken and one goes directly to the next operation after an indication of time corresponding to this last information. In the case of normal speed playback, successive bytes of the direct transcoded signal are examined successively at the rate of the MIC frames.

 The following operation amounts to determining the flow of the current time period; if the remaining duration is zero, the silence is ended, the current zero sample is in fact a linear sample of silence which is to be taken into account for the next phase SDL for reading the direct transcoded signal.

 If the remaining duration of the lapse of time is not zero, a steplength counter of the lapse of time is decremented by one step, initially loaded as a function of the current lapse of time and from the moment when a SY synchronization marker MIC is present. there was resumption of the SDS2 sub-phase before counting.

A second SDL phase corresponding to a reading of the direct transcoded signal is presented in FIG. 8
In the exemplary embodiment proposed, this SDL reading phase requires the reception of a synchronization reference point SY in order to authorize the potentially destructive reading process only from the moment when the receiving equipment is ready, however this is just one possible variant. This reception here triggers an acquisition operation of a word which is on the one hand immediately exploited, on the other hand temporarily kept for exploitation with the next word.

 Immediate processing consists in considering whether the received word which is called current is an extremump word, nor yes the indication of amplitude and the indication of duration which it comprises, are taken into account.

 If a rapid restitution of the initial signal is desired, the indications of time lapse of the extremum words are then modified using a specific conversion table so as to correspond to reduced time laps between extremas.

If the counting duration indication is zero, it corresponds
in the eode chosen at the first step of duration, the next phase SDE is a
word processing to provide a MIC sample corresponding to a
extremum of the initial signal.

On the other hand, if the counting duration indication is different from
zero, the direction of variation of the current sample is determined by
relation to the direction of variation defined for the previous sample and we
performs an SDI sample calculation phase using the ji samples.

previous and current extremum
If the word received after the acquisition operation is not a word
of extremum, one examines if it corresponds to an end of signal code. Such
end code triggers an SDF end of transmission process which will not be
not described further since it only has an Indirect report
with the method according to the invention.

If the word received after the acquisition operation is neither a word
extremum, nor an end code, it is recognized as the first byte
silence information via its descriptor and
it is loaded into the silence duration counter as well as the byte
acquired from his following.

An SDS2 sub-phase of silence processing is then triggered
from the silence information obtained.

The SDI sample calculation phase presented in Figure 9 concerned
MIC samples intended to be inserted between two samples
of successive extremum. The principle of the calculation consists in calculating the linear coded amplitude of the sample to be supplied by iteration from
amplitudes of the two extremas that surround it.

We therefore have the amplitude A. of a sample in linear coding from the amplitude of the previous Ai 1 by the formula
Ai = Ai-1 # integer value of # A
# T in which ss A is the difference in amolitude between the sample Ai
linear coded and the amplitude of the following extrema and T the time lapse
separating the sample Ai from the following extrema.

Having memorized two words of extremum and / or pseudo-extremum success
We therefore seek to determine what is the direction of variation which leads from the first to the second.

 For this purpose, the SDI sample calculation phase includes an operation of determining the direction of variation between successive extremum and / or pseudo extremum words.

If the direction of variation is positive, the samples inserted between the two extradas considered are determined by adding the correction factor corresponding to the integer value da + å the value
abut goal or deduced from the amplitude in linear coding of the previous sample. If the direction of variation is negative, the correction factor is of course subtracted from the value calculated or deduced from the previous linear coded sample.

 Then, in either case, the calculated amplitude is coded in logarithmic law A after any amplification or attenuation which is preferably obtained by passing the coding range indicated by the range change word N2 placed upstream of the first in time of the extras taken into account for the determination of MIC samples at the lower or higher coding range as the case may be.

 The choice of range and consequently of amplification is followed by an operation of recovery of the MIC synchronization reference mark St which results in the emission of a representative MIC sample Sun point of the initial signal and the decrementation of the time lapse duration counter. which defines the number of samples remaining to be supplied between two extremes etiou pseudo-extremas, after the generation of a MIC sample included between the ÉtIC samples corresponding to these extremes.

 If the remaining counting time is zero, we go to the SDE phase of sample coding corresponding to an extremum; if the remaining counting time is not zero, we start a new SDI sample calculation phase.

 The phase for determining the SDE extremum samples presented in FIG. 10 started with a search aimed at determining whether the word received is a range change word in view of its descriptor; If yes, the amplitude of the current word is determined both in the current range and in the future range which has just wandered indicated by the word change of range. The range change does not take place on the samples of the signal extrema, but on the extrema level.

 The next word is coded in current range and takes the place of the previous year memory An amplification or weakening of the samples can then be obtained in the manner already mentioned above, following this last possible operation, the MIC coded samples in law A are transmitted and a new direct transcoded signal reading phase is initiated.

Claims (10)

1 / Method of compression-decompression of a voice type signal coded in the form of a sequence of amplitude samples according to a technique known as pulse modulation and coding, characterized in that the samples corresponding to the extremes of the signal coded in binary words of extremum each comprising on the one hand an indication relating to the amplitude of an extremum, obtained from 11 corresponding sample, on the other hand an indication relating to the lapse of time separating this sample from 11 sample corresponding to the previous extremum and in that one transcodes the succession of samples corresponding to a temporary state of silence of the coded signal, following the recognition of this state, by binary information of silence, composed of at least one word and possibly repeated, which includes a descriptor and an indication relating to the duration of the silence, so as to reduce the volume of binary data necessary for the reconstruction of the sig nal. 2 / A method according to claim 1, characterized in that one identically transcodes to an eellantlllon corresponding to an extremum, so-called pseudo-axtremum samples which are respectively separated each from the sample corresponding to the extremum or to the pseudo- extremum which precedes it by a period of time equivalent to the maximum indication of the duration of a word of extremum. 3 / A method according to claim 2, characterized in that the samples which correspond to extremes translatable by a same indication of amplitude and which are separated from the extremum or from the preceding pseudoextremum by a period of time less than the maximum indication of duration of a word of extremum, are eliminated.  4i Method according to claim 1, characterized in that the generation of silence information (is triggered by the overflow of a duration counter incremented with each acquisition of sample in established silence phase and in that-the generation d2 an end of silence information (ES2) is triggered by exceeding a silence output threshold of a first bounded up-down counter incremented with a first step of incrementation by the samples whose amplitude is greater than an output reference value of silence (Af) and decremented with a first step of decrement lower than the first step of incrementation by the samples whose amplitude is less than the setpoint of silence output 5 / Method according to claim 49 characterized in that ~ l entry into the established silence phase, defined by a word of zero amplitude (MO), is triggered by exceeding a silence entry threshold (Es) of a second up-down counter limited in incremented with a second increment step by samples whose amplitude is less than a silence input setpoint (Ad) and disoriented with a second decrement step greater than the second increment step by samples whose l amplitude is greater than the silence input setpoint. 6 / A method according to claim 1, characterized in that the embers indexes of amplitude of words words of extremum and pseudo-extremum which are obtained from the corresponding samples are coded in one or the other of at at least two coding ranges, partially overlapping and individually selected as a function of the increasing energy of the coded signal which is measured from the samples. 7 / A method according to claim 6, characterized in that the energy measurement is carried out by rectification and low-pass filtering of the succession of samples corresponding to the initial coded signal. 8 / A method according to claim 6, characterized in that the conversion range changes as a function of the energy measurements are individually coded in the form of a range change word which includes a description and an indication relating to the chosen range and which is inserted before the extremum words transcoded using the range indicated. 9 / A method according to claim 1 or claim 8, characterized in that for the reconstitution of a coded signal after the previously claimed transcoding phase, the amplitude samples sought are obtained by a reverse transcoding phase of the amplitude indications. words of extremum and by a reverse coding after linear interpolation between extremas for the samples of amplitude to be inserted between words of extremes said linear interpolation being carried out starting from two pairs of calculation values resulting from the indications of amplitude and lapse of time determined from the indications relating to two successive words of extremum for the sample of amplitude preceding that which one calculates and for the second in time of the two words of extremum considered. 10 / A method according to claims 6 and 9, characterized in that the phase of direct transcoding of the amplitude samples to the extremum words and the phase of reverse transcoding of the extremum words to the a-amplitude samples s' eSfectute by means of transcoding tables which are individually assigned to the ranges in connection with the energy of the coded signal and which are deduced by simple homothety from one range to another. 11 / A method according to claim 10, characterized in that quton changes the decoding range from one range to the next to amplify or attenuate the coded signal. 12 / A method according to claim 9, characterized in that lon ensures rapid restitution of the significant constituents of the initial signal, in particular in the case of speech signals, by eliminating the silence information during the reverse transcoding phase.  13f A method according to claim 9, characterized in that one ensures a rapid restitution of the initial signal, in particular in the case of speech signals, by converting the time lapse indications of the extremum words into indications of shorter durations. during the reverse transcoding phase.