CH639218A5 - VOICE ENCODING / DECODING METHOD. - Google Patents

Abstract

In a voice coding system each vowel or consonant sound, or phoneme, is divided into repeated similar sections and the analog electrical equivalent of one of those sections is transmitted. If repetitive sections cannot be distinguished, e.g. during unvoiced consonants, a signal is derived from a section of predetermined length. The signal is transmitted in digital form together with signals indicating the number of sections in the phoneme, the intensity of the phoneme, the intensity of each of the repeated sections of the phoneme, any pre-equalisation applied to the speech, and phoneme change. On reception the analog electric equivalent signal is recovered, stored, level controlled, equalised, and repeated the appropriate number of times until a phoneme change is indicated.

Description

The present invention relates to a speech coding / decoding method.

It is often necessary to convert the analog electrical equivalent of the voice into digital form, e.g. for translations with computer etc., necessary. Because of the high bit rate of approx. 60,000 bit / s, which requires simple sampling and conversion into digital form, various coding methods for the voice have emerged (vocoder).

These methods are based on various principles, e.g. in a division of the analog electrical equivalent of the voice, which is abbreviated as AEÄS, into several frequency bands, e.g. 32 and coding the number of each volume and the envelope. During the restoration of the AEÄS, the intensity of a number of oscillators, according to the bands, is varied according to the abovementioned coding, and accordingly the composition which gives the AEÄS takes place. All known speech coding methods result in a bit rate of less than 6000 bit / s, but their practical implementation is very difficult and expensive.

The present invention is based on a method other than the known methods and takes advantage of the following property which the consonants and vowels of the voice have. The signal sections corresponding to consonants and vowels which make up the human voice and which are hereinafter referred to as phonemes have a signal form which consist of similar repetitive signal subsections which differ from the total intensity of each subsection. For some phonemes, e.g. ch and i | /,

the subsections are not repeatable in nature, but have a noise form. We also call the similar subsections groups, and in the case where the signal form of a phoneme is not repeatable in nature but has a noise form, we define as a group or Signal subsection, part of the signal section of a predetermined constant period of time. While speaking, each phoneme usually has a duration of 58 to 200 ms and each group a duration of 4 to 8 ms. The duration of the repeating groups resp. Signal subsections of a phoneme, respectively. Signal sections can also be less than 4 ms, if they consist of only one signal period, in exceptional cases they are also greater than 8 ms.

The groups resp. Sub-sections that make up each phoneme are similar for the same person and intensity range of speaking among themselves with a variation in their intensity. The intensity of the subsections within the same phoneme starts at a certain value, usually reaches its maximum value on the second or third repetition of the subsection, and then decreases towards the end of the phoneme, where it reaches a minimum value.

This knowledge now enables the implementation of a speech coding method, which is characterized by the characterizing part of claim 1.

The formation of the phoneme from similar groups enables a phoneme from the AEÄS of a subsection and intensity data of each subsection, e.g. to determine the maximum intensity, the integral of the intensity or a partial integral of the intensity.

According to the present invention, a subsection is scanned for each phoneme, the analog samples are converted into digital and the intensity of each group or. each subsection GID or certain groups digitizes GID. In addition, data regarding the duration of each subsection or the number of subsections of each phoneme as well as phoneme alternation characters are determined. The phoneme change character is a digital signal of a predetermined form which separates the data of each phoneme. During the decoding of the voice, the scanning data of the scanned subsection are continuously repeated within the duration of a phoneme, and after the conversion into analog form, the intensity is adjusted according to the intensity that each subsection had when coding. The resulting process is much simpler than the known one and can largely be implemented with logic circuits. The bit number of each scan is reduced by an intensity normalization of the voice, which is referred to hereinafter as compensation, in one or two steps. In the case of two stages:

a) The AEÄS is partially integrated over a period of several phonemes, which is called the formation of long-term intensity.

b) By means of an amplifier with digitally variable amplification, the intensity of the AEÄS is multiplied or divided by / by the digital value of the long-term intensity of the voice, so that an AEÄS / LZ is created, the long-term integral having a constant value during the time of several phonemes.

The AEÄS / LZ, which is the result of multiplication or division with the long-term intensity, is integrated during the duration of a phoneme, the result is called the phoneme intensity.

d) The intensity of the AEÄS / LZ is multiplied or divided by / by the digital value of the integral during a phoneme, i.e. the phoneme intensity, by a second amplifier with digitally variable amplification, whereby the AEÄS / Ph is produced, the phoneme intensity being in the interval of one Phonems has a constant value.

In the case of a stage, a) and b) are omitted and the AEÄS / Ph signal is formed directly. The integral over a phoneme required for this can be replaced by another intensity variable. In order to obtain the same form of results in one stage, the digital implementation of the integration result in one stage must have a number of bits which is equal to the sum of the number of bits obtained in two stages. In both cases, the scanning is carried out by the AEÄS / Ph signal.

The code of each phoneme consists of:

a) the phoneme change character, which e.g. consists of nine «1» and one «0» or nine «0» and one «1» (8-10 bits per phoneme).

b) the digital data of the phoneme intensity PhID (4 bits per phoneme) for an equalization level or the digital data of the long-term intensity (2-3 bits per phoneme) and the digital data of the phoneme intensity PhID (1-3 bits per phoneme) with two equalization levels.

c) the data of the number of samples of a subsection or subsection number of the phoneme (6 bits per phoneme).

d) the digital data of the samples of a subsection (4 bits per sample or 230 bits per phoneme for 8 ms groups).

5

10th

15

20th

25th

30th

35

40

45

50

55

60

65

639 218

4th

e) the data of subsection intensity GID (2 bits per d) the digital data of the scanning of a part of the

1 or 2 groups). Subsection within a certain time (e.g. 2 ms)

The specified number of bits for the above-mentioned data for area II. The number of samples is always fixed. Codes are purely indicative and can be determined depending on the one required (e.g. 12-16 bits).

Quality of the voice and the maximum permissible frequency 5 e) the data of the partial integral of the intensity of the others are changed. Parts of the scanned subsection. The number of these parts

The data of the encoded phoneme are always stored in series, e.g. 4, fixed and ordered for smaller subjects and with a certain delay, e.g. sections with the non-existing parts as existing

200-300 bits, saved or viewed for further processing or intensity 0.

Transfer forwarded. to f) the digital data of the sampling of a sub-

An order is first cut for the decoding of the voice, e.g. the fourth, in area I (4 bits per scan of the coding data and their entry in or for group length 8 ms approx. 58 bits).

speaking memories or shift registers. g) the data of the intensity of the subsections of the

The digital data of the scanning of the subsection area I and the area II (2 bits for each area are delivered in continuous repetition, as long as 15 subsections or two subsections each).

is the duration of the phoneme. Through a digital / analog data a) to e) always have constant bit numbers. The

Transducers, further abbreviated to DAC, these data compensations to obtain the AEÄS / Ph signal can be converted into the AEÄS / G of the subsection, which was also sampled before the division of the frequency band into two areas, which is only within a subsection, with the reverse compensation or. Modulation exhibits changes. 20 when decoding according to the composition

Via an amplifier whose amplification must be digital.

according to the data of the respective subsection intensity The data of the coding are stored in the memory as follows

changes, the AEÄS / G is divided or multiplied so that:

an AEÄS / Ph arises, which changes only within the first the phoneme change character and then all data, intervals of a phoneme. The AEÄS / Ph is always led to the input with fixed bit numbers, from b) to e), after the same gear of a second amplifier, the amplification of which follows the order. The order of the order is irrelevant to the digital data of the phoneme intensity PhID. The scan data of area I follow, e.g. of the four or is multiplied, and in the case of a stage the th subsection and finally the data of the subsection

original AEÄS results. In the case of two stages there are cutting intensities. The data are presented with a minimal AEÄS / LZ, which is output via a third, with the first 30 delay m between 100 and 300 bits, and second amplifier-like amplifier by dividing. The decoding of the voice for area I is done as or multiplying that original AEÄS results. The travel is described on page 5, line 18 to line 28. When ordering the reverse equilibrium resp. However, modulations of the data become those of area I, of which the three AEÄS does not matter, and e.g. only the second, area II separately according to the order in which they are then used in the first amplifier. 35 of the coding were ordered. Also at the

A further reduction in the number of bits per group subsection intensity the first two bits for the one. Sub-section, ie also the number of bits per phoneme, frequency range and the second two bits for the other can be achieved by providing the frequency band of the voice frequency range. For area II they are divided into two areas and each area is coded separately. Sampling data of the first part of the subsection in continuous. The decoding is likewise carried out separately for each area 40 repetitions, as long as the phoneme lasts and over and there follows the composition of the two areas to the one DAC converted into analog form. This is followed by the reverse AEÄS / Ph after the digital / analog conversion and after the return compensation by an amplifier with controllable compensation. The coding of the voice in this case is strengthened, according to page 5, line 30 to page 6, line 9, as follows: according to the data of the other parts of the subsection. The-

The frequency band of the voice is equalized by two filters in 45 ser inverse compensation. this modulation is performed for area I, which repeats the low frequencies under each subsection, using the information on

1 / b passes fmax, where most suitable b between 3 and 5 is the subsection length given by area I. That and divided into area II, where the frequency 1 / b fmax to fmax by AEÄS / GII, which results from this compensation, will be left. The area I is coded as described on the page using a second amplifier for the signal AEÄS / Phll according to 3, line 1 to page 3, line 31, but 50 the two remaining bits of the subsection intensity are compensated with sampling intervals that are b times larger . In area II according to the chen resp. modulated as in area I. By a third equalization according to page 3, line 33 to page 4, line 19, the AEÄS / Phll signal becomes the original subsection, which was sampled in area I, in time AEÄS des Area II balanced or modulated. Divided by intervals of approx. 2 ms and the first time interval with a composition of the AEÄS signals of the two areas is sampled in a sampling time somewhat less than 1/2 fmax, the sampling 55 finally forms the original AEÄS signal. The bit values are then digitized. By means of a system of partial integers per phoneme, approximately half of the bits required are in opposition to the remaining parts of this subsection and implemented via the coding in a frequency range.

The value of the integration into digital form is then explained. The invention is then explained, for example, using the data of the remaining time intervals of this subsection of two figures.

receive. so it show:

In the case of the two frequency ranges, the code of FIG. 1 consists of a speech code phoneme based on a block diagram: device for explaining the method,

a) the phoneme change character (8-10 bits). Fig. 2 is a block diagram showing a device for b) the group length, i.e. the data of the number of sampling speech decoding according to the method.

of the subsection in area I or the number of subsections. In FIG. 1 there is an example of a coding method for the sections of the phoneme (6 bits). Voice (vocoder) according to the present invention c) the data of the phoneme intensity for each range (4 in a frequency range and with an equalization level bit). represented by a block diagram. The integrator Ini

5

639 218

makes a partial integration over the first (m - 1) subsections of the AEÄS signal, the result is stored via an analog-to-digital converter ADC2 for the duration of a phoneme. This value (phoneme intensity or signal section intensity) is used to digitally adjust the gain of an amplifier VI, at whose input the AEÄS signal is fed. By changing the gain, the AEÄS / Ph signal is generated at the output of the amplifier VI, the normalization variable, the result of the integration at Ini during the time of a phoneme, has a constant value.

A low-pass filter F1 at the output of the amplifier VI does not let through the frequencies which are greater than the maximum frequency fmax which is contained in the coding. An AD converter ADC1 converts the AEÄS / Ph signal at the output of the low-pass filter F1 into digital form. At time intervals Td, which are to be regarded as sampling intervals, the instantaneous values at the outputs Q1 ... Q4 of the AD converter ADC1 are stored in a shift register SRI via the inputs 1 ... 4. The time Td is slightly less than 1/2 fmax. An integrator In2 integrates the AEÄS / Ph signal over one or two subsections, the result is converted into digital form by another AD converter ADC3 with parallel outputs Q1 and Q2 and each for one or two subsections in the shift register SR2 whose parallel inputs 1 and 2 are stored for the corresponding time (subsection intensity data GID).

A circuit GW for indicating the change of subsection generates a pulse G-I at the beginning of each subsection. If no subsections are recognizable, this circuit GW generates a pulse at certain time intervals, e.g. between 8 and 9 ms (see page 2, line 12). If the repeating subsections are less than 4 ms, the circuit GW generates a pulse for every second subsection. The change of subsection is recognized by a charging current pulse on a capacitor when it is charged via a diode of AEÄS / Ph signal. This capacitor is easily discharged through a resistor during each subsection. The pulse G-I always coincides with the beginning of a period TE of the output system. The change in subsection is much more clearly recognizable if it is detected by the AEÄS signal via an amplifier, the amplification of which is varied analogously with the input voltage, or via an exponential amplifier. In this case it is expedient to use the maximum value of the last-mentioned signal instead of the integral for the compensation of the AEÄS signal.

A phoneme change circuit PhW in connection with the circuit GW and a central processing unit CUI (see page 13, line 21; page 14, line 8) emits a phoneme change pulse Ph-I with each phoneme change. The function of this circuit is based on the correlation between two successive groups. Subsections. The sample data of each subsection from the shift register SRI appear at its outputs Q1 ... Q4 and are loaded at inputs DI into the memory of the circuit PhW until the correlation with the sample data of the next group or. of the next subsection. For correlation, the analog value of each scan of a subsection is multiplied by the digital value of the corresponding scan of the previous subsection that was the same amount of time from its beginning. This multiplication is carried out by a digitally controllable amplifier, the gain of which is changed by the data in the memory of the circuit PhW. Then these product results are added and the total result divided by the number of products, which gives a maximum value as long as the subsections are similar. A sudden decrease in the total value indicates a phoneme change, whereupon the phoneme change pulse Ph-I is generated. The Ph-I pulse is thus generated at the end of the first subsection of a phoneme when the correlation between it and the last subsection of the previous phoneme has ended. The phoneme change pulse Ph-I is always generated together with a subsection change pulse G-I.

The scanning data of the nth subsection (e.g. n = 4) are entered by the serial output Q1 of the shift register SRI into a memory Ml (e.g. 256 bit) via its data input DI and the number of scans on this subsection is kept at the central unit CUI (see Page 13, line 22) and during the change of the subsection to n + 1 at the first parallel inputs DI of a shift register SR3. The next inputs DI of the shift register SR3 are connected to the parallel outputs Q1 ... Q4 of the memory circuit 4FF1 in order to enter the phoneme intensity. The following input is connected to - Vss of the power supply and the other inputs to + VDD of the power supply in order to enter the 8 to 10 bits for the phoneme change character. If the feed clock pulses Tj of the shift register SR3 start with a delay, the number of its inputs which are set to “1” for the phoneme change character can be reduced accordingly.

All data that are entered in the shift register SR3 and in the memory M1 and that which are entered after the subsection n in the shift register SR2 are stored in the memory M2 at the beginning of the n + 1th subsection and until the end of the phoneme (for example 512 bits ) written. The data of the following phonemes are entered successively in the memory M2 until it is full, after which the remaining entries in the cycle are continued from the beginning by deleting the previous data. An electronic switch ES connects the n + 1-th subsection from the data input DI of the memory M2 first to the output DO of the shift register SR3, then to the data output DO of the memory Ml and finally to the data output DO of the shift register SR2 and in succession the phoneme change character Enter the data of the phoneme intensity, the number of samples which has a fixed number of bits, then the sample data of subsection n and finally the subsection intensities.

All control of the entry of the data for phoneme change, phoneme intensity, number of samples, sample data and subsection intensity data in the memories 4FF1, Ml, the shift registers SRI, SR2 and SR3 and subsequently in the memory M2, as well as the various clock pulses Cl and the programmed pulses T; For the input and output of the data on the various circuits and the counting of the subsections, is carried out by the central processing unit CUI. The central unit CUI also specifies the addresses for writing and reading in the memories M1 and M2. The central processing unit CUI has a counting unit which controls the difference between the read addresses and the write addresses (see page 5, lines 13-16 to page 14, line 14).

The central unit CUI can comprise a microprocessor or consist of an integrated circuit of an oscillator, counters, a presettable up and down counter and gates, which generate the necessary commands in a known manner. The pulses G-I and Ph-I give the central unit CUI the group change resp. Subsection change and phoneme change information.

The memory M2 is read at the frequency fE, which is the data output frequency. Reading always takes place between two write operations, so that write and read commands are never executed simultaneously. Reading takes place with a delay of more than m bits from writing

5

10th

15

20th

25th

30th

35

40

45

50

55

60

65

639 218

6

(e.g. m = 256 bit) instead. The central unit CU 1 outputs the read addresses to the memory M2 during the read time, and the write addresses during the remaining time.

The output control AR comprises a flip-flop, a shift register and an 8 input NAND gate and outputs the data of the memory M2 in series at the data output SDO of the system. When there is a change of phoneme with the seven consecutive “1”, the output control AR sends a “1” signal to the central unit CUI. If the difference between the read addresses and the write addresses is less than m, then this “1” signal from the output control AR via the central unit CUI stops reading the memory M2 until the difference becomes greater than m again, while the output control AR continuously “1 »Gives up until reading starts again.

In the case where the central processing unit CUI comprises a microprocessor with a word length a bit, the AD converters ADC1 and ADC2 are a-bit converters and the AD converter ADC3 are a / 2-bit converters, the memories M1 and M2 be organized with a parallel 1 and outputs and the output control AR will have an additional shift register instead of the flip-flop, with a parallel inputs and a series output for the series output of the data of the system.

Figure 2 shows a block diagram of a corresponding decoding system. The data of the encoded language are first fed into a shift register SR4, which detects the phoneme change character. If there are more than seven consecutive "1" and then the first "0" occurs, this is the beginning of a new phoneme or. Interpreted signal section. The section intensity data and the subsection length data that follow the “0” are entered in a shift register SR5. Then the sampling data, the number of which is given by the subsection length data in the shift register SR5, is alternately for each phoneme. each signal section is written in memories M3 and M4. After the completion of the writing of the number of samples, the subsection intensity data are alternately entered in two memories M5 and M6 until the phoneme change sign appears. The entry via a shift register SR6 creates a delay of seven bits, which prevents the entry of the first seven «1» of the phoneme change character. The memories M3, M4, M5 and M6 can be separate memories, or a memory for M3 and M4 and a memory for M5 and M6 can be provided, where M3, M4 or M5 and M6 are parts of a memory with specific addresses.

After entering the scan data, the section resp. Phoneme intensity data into a data storage circuit, e.g. a multiple flip-flop 4FF2 rewritten and the data of the number of samples stored in a data storage circuit or set on a presettable up and down counter within the central unit and until the data of the next section or. Preserve phonems. If the sample data in the memories M3 or. M4 are entered, reading begins in successive cycles. The number of samples stored in the shift register SR5 gives the central unit io CU2 (see page 17 from line 14) the data for the length of the repetition of the cycle mentioned. The output of the memory M3 or M4 gives, in parallel, the digital data of each scan to a data storage circuit 4FF3 via the shift register SR7. A DA converter DAC1 converts the digital data 15 of the samples into analog signals, and the AEÄS / G signal of the subsection n thus created is repeated for each sampling cycle.

After a low-pass filter F2, the analog signal AEÄS / G is led to the input of an amplifier V2, the 20 gain of which is digitally changed by the data of the subsection intensities, which are entered in the memories M5 or M6. A shift register SR8 outputs the data of the respective subsection intensity in parallel to a circuit 2FF1, which keeps them for the duration of one or two subsections. At the output of amplifier V2, the AEÄS / Ph signal appears, which is routed through a second amplifier V3, the amplification of which is digital by the phoneme or. Section intensity data is changed, which gives the original AEÄS signal. A second low-pass filter 30 ter F3 attenuates the higher frequencies, which can occur due to the rapid changes to the amplifiers V2 and V3. In the case where the memories M3, M4 and M5, M6 have parallel outputs, the shift registers SR7 and SR8 are omitted.

35 The whole control of the entries of the section intensity data, the number of samples, the alternating entries in the memory M3 or. M4 and M5 resp. The central unit CU2 executes M6, as well as the various clock pulses Cl and the programmed pulses Tj for the input and output of the data on the various circuits. The central unit CU2 gives the addresses 0 ... 7 and 0 ... 6 to the memories M3 and M4 as well as M5 and M6. It can comprise a microprocessor or consist of integrated circuits of counters, shift registers, flip-flops and 45 gates, which deliver the necessary commands in a known manner. As is known, the oscillator which gives the clock pulses is synchronized with the frequency of the input signal.

2 sheets of drawings

Claims (9)

639 218 PATENT CLAIMS 1. speech coding / decoding method, characterized in that for coding a) an analog electrical speech signal (AEÄS) is generated by a converter, divided into temporal signal sections, corresponding to phonemes (PhW), ai) it detects repeated signal subsections of a similar shape, from which the phenomena signal sections are composed (GW), a2) if no repeating signal subsections can be detected therein, signal subsections of a predetermined time length are defined, b) measures the duration of the signal subsections for each signal section, digitizes and stores the measured value (CUI, SR3), c) signal subsection-specific intensity values (Gl) measures (In2), the measured values are digitized (GID) (ADC3) and stored (SR2, M2), d) measures a signal section-specific intensity value (Phl) (Ini), digitizes the measured value (PhID, ADC2) and stores it (4FF1, SR3), e) selects at least one signal subsection per signal section and normalizes its intensity with the signal section-specific intensity value (VI), samples the standardized intensity at a predetermined rate and stores the digitized (ADC1) samples (MI), f) compiles the digitized, signal section-specific intensity value (PhID), the digitized time duration value, the digitized sample values and the digitized subsection-specific intensity values (GID) for each signal section in a predetermined order, between two signal section limit signals in each case. 2. Speech coding / decoding method according to claim 1, characterized in that one identifies the individual signal sections and generates a signal section limit display signal (Ph-I), preferably as a pulse, in that the identity of each signal subsection with the signal section -specific intensity value (Phl) normalized, samples and stores the normalized intensity at a predetermined sampling rate (PhW), which preferably multiplies digital samples of each signal subsection by the corresponding, preferably analog samples of the next signal subsection, adds the resulting product values of each signal subsection, successively Compares pairs of the sum values and generates the signal section limit indication signal (Ph-I) when the comparison satisfies a predetermined inequality. 3. The speech coding / decoding method as claimed in claim 1, characterized in that individual signal subsections are identified by detecting each capacitor charging current pulse on a peak-indicating diode capacitor circuit (GW), measuring the time interval between subsequent charging current pulses and a signal subsection limit display signal (GI) preferably generated as a pulse when a charging current pulse occurs when the time interval lies between a first and a second predetermined limit value, otherwise either the signal subsection limit display signal is generated in a predetermined time sequence if the time interval exceeds the second predetermined limit value, or when Occurrence of every second charging current pulse is generated when the time interval is less than the first value. 4. speech coding / decoding method according to claims 2 and 3, characterized in that one controls the compilation of the digitized values by means of a central processing unit (CUI) and the latter by means of the signal section limit display signals (Ph-I) and the signal subsection Limit display signal (GI) time-controlled. 5. Speech coding / decoding method according to claim 2, characterized in that for each signal section the digitized samples of the selected signal subsection are loaded serially or in parallel into a first memory (Ml), after sampling the signals for the signal section limit, the time period and enters the digitized signal-specific intensity signal (PhID) into a shift register (SR3), writes it from the shift register (SR3) into a second memory (M2), then the samples from the first memory (MI), then with a predetermined delay, the digitized Sub-section-specific intensity values (GID in SR2) writes to the second memory (M2). 6. speech coding / decoding method according to claim 5, characterized in that one writes the data for the signal sections serially in the second memory, it cyclically renewed at full memory between write operations in this memory, and in this regard, reads output data from this memory (M2) , and if a certain delay between the read operation and the write operation is undershot, when a signal section limit signal is detected, the reading is made impossible until the specific delay is reached and then released again. 7. Speech coding / decoding method according to claim 1, characterized in that for the decoding a) signal section boundary signals are detected, these and the section specific digitized intensity values (PhID), the duration data, the sample values and the subsection stores specific digitized intensity values (GID), b) the samples are repeated in an ordered cycle, converted into analog form, which results in repetitive subsection signals with standardized intensity, in accordance with the time duration data, c) modulating (V2) the analog subsection signals with the subsection-specific intensity values (GID) to produce signal sections with normalized intensity, d) the signal sections are modulated with the signal section-specific intensity values (PhID) in order to regenerate the speech signal (AEÄS). 8. Speech coding / decoding method according to claim 1, characterized in that the analog electrical speech signal (AEÄS) is divided into two frequency ranges from zero to f "and fn - fmax with preferably fmax / fn = 2 to 5 and the coding of each frequency range ( I, II), whereby in the higher-frequency range (II) the subsection to be scanned is divided into several time segments, the scanning is carried out only on the first segment with a predetermined number of scans, and segment-specific intensity values are determined for the remaining time segments and the data of the one range (II), with a predetermined constant number of bits, combined with those of the coding of the other area (I), preferably precedes the data of one (II) with that of the other (I). 9. speech coding / decoding method according to claims 7 and 8, characterized in that for the decoding a) preferably by means of shift registers, the coded data of one area (II, f "- fmax), which assume a constant number of bits, of the Data from the other area (I, f0 - fm) separates, b) the data of the two areas (I, II) are stored separately, those of the other area (I, f0 - fn are decoded, c) the sampling data of the first time segment is repeated cyclically for the duration of a signal section, in analog 2nd 5 10th 15 20th 25th 30th 35 40 45 50 55 60 65 3rd 639 218 Form changes so that repetitive, analog time segment signals are created with standardized intensity, d) these analog time segment signals are modulated with the intensity data of the other time segments, with which standardized subsection signals are created, which are then modulated with the subsection-specific intensity values, with which standardized signal sections are created in a region (II), and these normalized signal sections are finally modulated with the signal section-specific intensity value , which creates the signal section in one area (II), 0 composes the signal sections of the two frequency ranges (I, II) to obtain the regenerated speech signal.