GB2233195A

GB2233195A - Speech signal transmission

Info

Publication number: GB2233195A
Application number: GB8114115A
Authority: GB
Inventors: Rudiger Reisse; Gerhard Wimmer
Original assignee: Siemens AG
Current assignee: Siemens AG
Priority date: 1980-06-09
Filing date: 1981-05-08
Publication date: 1991-01-02
Anticipated expiration: 2001-05-08
Also published as: PT73132A; GB2233195B; IT8121545A0; DK241181A; IT1228133B

Abstract

In a vocoder speech transmission process a synthesiser component (ST') at the receiving end has no spectral channel filter, but simulates its pulse signal responses and noise signal responses (hk(i), rk(J)), which are respectively stored in separate read-only memories (ROM H, ROM R), and read-out from each of these read-only memories is effected during a synthesis interval in dependence upon the tuning character criterion (Sc) for the weighting of the signal responses with the associated transmitted envelope curve values (ak), and the signal responses, together with the envelope curve values, are fed to a multiplier-adder. <IMAGE>

Description

METHODS OF SPEECH SIG%L. XENSMISSION The invention relates to methods of speech signal transmission, wherein at the transmitting end of a channel a speech signal is fed to an analysis component of a vocoder for the acquisition of the speech-specific parameters, namely the fundamental speech frequency, a tuning character criterion, and envelope curve values for a specific number of spectral channels, each of whose frequency position and width is predetermined by a filter bank, these parameters being assembled in a suitable manner and transmitted to the receiving end of the channel in the timing of consecutive analysis intervals, and at the receiving end the transmitted parameters control, in the assigned spectral channels of the synthesis component of a vocoder, in dependence upon the tuning character criterion the amplitude of a pulse signal influenced in pitch by the fundamental speech frequency or the amplitude of a noise signal, and wherein subsidiary speech signals generated in this way are subsequently added to form a synthesised speech signal.

Known vocoder transmission processes of this kind are described, for example, in publications by J.L.

Flanagan in "Speech Analysis, Synthesis and Perception" 2nd edition, published by Springer Verlag, Berlin, Heidelberg, New Yorke 1972, pages 323 to 330, and in Siemens Zeitschrift, 47 (1973) 3, pages 131 to 135. As a rule the vocoder used in these cases employs the same band filters for synthesis as for analysis. Tn the case of speech transmission in semi-duplex-operation this results in a favourable1 outlay1 since one filter bank can be switched over, and consequently only one bank needs to be provided for each device. The band filters are normally constructed as passive L-C filters, which require a considerable space for installation. For mobile devices, which basically require to be of low weight and small volume, the space required for the aforementioned filters is frequently not available.

In place of passive L-C filters it is possible to use "SWITCHED CAPACITOR" filters, which are available in integrated form, and which require a relatively small installation space. However, like all scanning filters, these filters have the disadvantage that their transmission function periodically sweeps through the frequency range.

In other words, the input signal for such a filter must not comprise any spectral components with a frequency value above half the scanning frequency. Practice has indicated that the upper band limit of the input signal must possess an even greater interval below half the scanning frequency if disturbances are to be effectively avoided. This does not give rise to problems as regards the analysis component, as this interval can safely be maintained by simple means.

However, in the synthesis component this requirement can only be assured by a substantial additional financial outlay, since the excitation function for speech signals is a rectangular pulse sequence. In other words substantial analogous filter means must be used in order to effectively suppress disturbing effects in the form of a chirping background noise.

Regardless of the type of filters used in the spectral channels of the synthesis component, a fundamental disadvantage results from the use of such filters. In order to achieve as flat a spectral curve of the excitation spectrum as possible, the excitation pulse must be relatively short. The energy content of this excitation pulse is then correspondingly low, so that its harmonics, which excite the filters, exhibit only a small amplitude, and thus cannot fully modulate the filters. The outcome is a relatively poor signal-to-noise ratio.

One object of the present invention is to provide a method for the transmission of speech signals of the type described in the introduction, providing a way to reduce the space requirements for the requisite filter means, but avoiding the difficulties involved in the use of scanning filters, and moreover facilitating an optimum signal-to-noise ratio for the speech signal generated in the synthesis component.

The invention consists in a method df transmission of speech signals from a transmitting end to a receiving end, wherein at the transmitting end the signal is conducted to the analysis component of a vocoder for the acquisition of speech-specific parameters, composed of the fundamental speech frequency, the tuning character criterion, and the envelope curve values of a specific number of spectral channels, whose respective frequency positions and band-widths are predetermined by a filter bank; wherein these parameters are assembled and transmitted to the receiving end in the timing of consecutive analysis intervals, wherein at the receiving end, a synthesiser component is provided in a vocoder, in which for each of the assigned spectral channels the transmitted parameters control the amplitude of a pulse signal which is influenced in pitch by the fundamental speech frequency or of a noise signal, in dependence upon the tuning character criterion; and wherein the subsidiary speech-signals generated in this way are subsequently added to form the synthesised speech signal, the pulse signal responses and the noise signal responses of spectral channel filters are simulated in the receiving end synthesiser component, each being stored in a respective read-only memory, weighted with the respective envelope curve values assigned to the spectral channels for all the spectral channels within one synthesis interval, being read out from one or the other read-only memory in dependence upon the tuning character criterion, and the number of the scan values of a pulse signal response which are to be read-out is effected in dependence upon the fundamental speech frequency when the read-only memory which contains the pulse signal responses is interrogated.

The invention is based on the fact that the synthesis component of a vocoder consists of a weighted summation of the outputs of a plurality M of band filters Bk. The respective envelope curve weights ak are set in accordance with the amplitude values and envelope curve values, determined in the analysis component, for each of the M spectral channels. The band filters are excited by pulse signals or noise signals e(t), and from their outputs supply pulse responses hk (t), and noise responses rk (t).If the synthesis becomes a time-discrete process at times t = i1t, the synthesised speech signal is represented by:

for speech signals and for other sounds by

In other words, the synthesis can be carried out, dispensing with filters, if the pulse responses hk(i) and the noise responses rk(i) of the spectral filters are stored in readonly-memories and, for the requisite multiplication with the envelope curve values ak, are read out from one or the other read-only-memory during a synthesis interval. The summation of the products then leads to the synthesised speech signal ~(~)in the desired manner.

By means of integrated technology this process can be used to design circuits which require an extraordinarily small installation space. The chirping background noise which normally occurs when scanning filters are used can be fully avoided. Furthermore noise-free signal responses can be obtained to be stored in a measuring arrangement designed specially for this purpose, so that in practical operation an optimum signal-to-noise ratio of the synthesised speech signal s(i) is assured.

In the implementation of a method in accordance with the invention, the transmitted speech-specific parameters are expediently firstly transferred, in the timing of consecutive analysis intervals, into a working store which forwards the envelope curve values to the intermediate store of a multiplier-adder, and forwards the parameters relating to the fundamental speech frequency and to the tuning character criterion to corresponding control inputs of a control unit.

During a synthesis interval, and in dependence upon the parameters with which it is supplied at its input, the control unit controls the read-only memories, the multiplier-adders, and the connection of the particular read-only memory whose content is to be read out to the multiplier-adder.

The conditions are particularly simple if the transmission of the speech-specific parameters is carried out in coded form in a time multiplex frame governed by the analysis interval, and the speech synthesis is carried out in digital mode at the receiving end. At the output of the multiplieradder, the digital synthesised speech signal is converted into an analogue mode via a following digital-analogue converter, which is likewise controlled by the control unit.

The invention will now be described with reference to the drawings, in which: Figure 1 is a block schematic circuit diagram of one exemplary receiver arrangement for a transmission channel, with details of the synthesiser component of a vocoder; Figure 2 is a graph showing the amplitude variations of a pulse signal response, plotted against time; Figure 3 is a graph showing the amplitude variations of a noise signal response; Figure 4 is a block schematic circuit diagram of one exemplary embodiment of a synthesiser component suitable for use in accordance with the present invention; and Figure 5 is a graph showing the amplitude variations of a speech signal which has been synthetically produced in the synthesiser component shown in figure 4.

The receiver arrangement shown in figure 1 receives an input signal se(t) in the form of a multiplex frame whose frame period corresponds to one transmitting-end analysis interval, and which contains within one frame period a coded form of the speech-specific parameters which occur in one analysis interval.

Following conversion in a digital-to-analogue converter D/A, the speech-specific parameters are distributed by a demultiplexer DE-MUX between respective inputs of a synthesiser component ST. The synthesiser component possesses M spectral channels, which each consist of a series combination of a respective multiplier, M1 to Mk, MM, and a respective band-pass filter, B1 to Bk, BM, each supplied at its input by an associated output of the demultiplexer with a selected envelope curve value, a 1 to ak, a M, which represent the individual weightings.The synthesiser component ST also possesses a separate input for a tuning character criterion Sc and another for the fundanental speech frequency, represented by the ratio No of the fundamental speech frequency period to the scanning interval, which is predetermined by the transmitting-end coder.

The synthesiser component ST also comprises a pulse generator PG, and a noise generator RG, the pulse generator PG being controlled in pitch by the input parameter No. The outputs of the pulse generator PG and the noise generator RG are selectively connected to the second inputs of the multipliers M1 to Mk, MM via a changeover switch U, in dependence upon the tuning character criterion Sc. Depending upon whether the excitation function e(t) for the spectral channels is a noise signal or a rectangular pulse signal, the signal responses consist of the respective pulse signal responses, generally characterised as hk(t), or the noise signal responses, characterised as rk(t). These signal responses are combined in an adder Su, to form a synthesised speech signal s(i) at the output of the synthesiser component ST.

In the graph shown in figure 2, the amplitude X of any particular pulse signal response hk(t) is plotted against time t, and in figure 3 a representative noise signal response rk(t) of a band-pass filter such as Bk is shown.

As stated above, the synthesised speech signal s(i) can be represented in accordance with equations (1) and (2) in the form of a weighted summation of the outputs of all the M band-pass filters. As the excitation function e(t) of a rectangular pulse sequence ox of white noise are known, the signal responses of the band-pass filters can also be given and stored in the form of scanned values in interrogatable read-only memories. Thus, provided there is access to the signal responses, in the synthesiser component ST it is possible to dispense with actual band-pass filters.

The weighting of the signal responses can be represented in matrix form as follows: H.A=S; (3) and R . A = S ; (4) with:-

h11 h12 h13 hlM h21 ..

H = h31 .

. . . .

hNl hNM and with:

rll rl2 rl3 rlM r21 R = r31 . .

rN1 r NM rN1 rNM where:- A = (a1, a2, a3 ... aM)' and S = (s1, s2, s3 ... sN)', Thus, as indicated by equations (3) and (4), the weighting of the signal response envelope curve values and their subsequent summation to form the synthesised speech signal directly leads towards a digitally operating realisation of the synthesiser component, in which the band-pass filters are not required; as synthesis is achieved from stored pulse responses and noise responses. The pulse responses and noise responses are multiplied with the relevant vector A in a synthesis interval Tms.As has already been noted, the number No of the scan values in the pulse responses is a function of the fundamental speech period duration To, which itself varies with time, and satisfies the equation: No = To(t)/ t (5) where the scanning time interval isss t.

In practice this means that, in respect of each pulse response, a read-only memory which contains the pulse signal responses must store a maximum number No-max of scanned values, which corresponds to the ratio of the maximum possible fundamental speech period To-max to the scanning time intervalAt, rounded up to a whole number. Accordingly, with an increasing fundamental speech frequency, during one synthesis interval an inversely proportional number of scanned values are read out from the read-only memory which contains the pulse responses, and in this way the pitch of the transmitted speech which is to be obtained by the synthesis is taken into account.

The exemplary synthesiser component ST' that is illustrated in figure 4 serves for the implementation of the method proposed in accordance with the invention, and is provided at its input end with a working store RAM, into which the speech-specific parameters are input in the timing of the frame period of the incoming t.d.m. frame.The parameter No which corresponds to the fundamental speech frequency, and the tuning character criterion Sc are fed via separate outputs to respective control inputs of a control unit SS whose output feeds a data bus C, and so controls a read-only memory ROM H that is provided for the pulse signal responses hk(i); a read-only memory ROM R that is provided for the noise signal responses rk(i); a multiplier-adder MA, and also a digital-toanalogue converter D/A connected to the output of the multiplieradder, and providing the output s(i).The working store RAM feeds all the envelope curve values such as ak to an intermediate store SR, which is assigned to the multiplier-adder MA, and across which the multiplication in accordance with equations 3 and 4 is effected; in that by means of the tuning character criterion Se currently available in the control unit SS data is either read-out from the read-only memory ROM H or the readonly memory ROM R, and the envelope amplitude values such as ak of the associated synthesis interval Tms, which are assigned to the read-out scan values of the signal responses multiplied, and the obtained products are subsequently added.

As regards this multiplicaton it should be assumed that the frame period Tm which agrees with the duration of one transmitting-end analysis interval Tma, will only by chance represent a whole multiple of the fundamental period duration To. Consequently, in the synthesis, during the multiplication of the pulse signal responses hk(1), measures must be taken to allow for these conditions by extending or shortening the synthesis interval Tms relative to the analysis interval Tma by a maximum of one fundamental period To.This is carried out in accordance with the following equation: Tms = Tma + To/r (6) where r is a shortening factor which is automatically taken into account by the control unit SS, in that the transition between two consecutive synthesis intervals Tms always coincides with the transition between two consecutive fundamental speech frequency periods.

For completeness the graph shown in figure 5 represents the amplitude X of the synthesised signal s(i) which has been produced in the arrangement shown in figure 4, plotted against time t.

For the construction of the synthesiser component ST' shown in figure 4, the multiplier-adder MA can consist of a high-speed multiplier of TTL design, for example, as manufactured under the designation NPY 8HJ-1 by TRW-LSI Products. Eight-bit PROMS having a storage capacity of 16 Kbit are suitable for the read-only memories.

At the read out of the noise responses from the read-only memory ROM R, each noise signal response is expediently represented by a fixed number of scan values, which are obtained from the ratio of the frame period Tm to the scanning time interval bt.

Claims

Claims: -

1. A method of transmission of speech signals from a transmitting end to a receiving end, wherein at the transmitting end the signal is conducted to the analysis component of a vocoder for the acquisition of speech-specific parameters, composed of the fundamental speech frequency, the tuning character criterion, and the envelope curve values of a specific number of spectral channels, whose respective frequency positions and band-widths are predetermined by a filter bank; wherein these parameters are assembled and transmitted to the receiving end in the timing of consecutive analysis intervals; wherein at the receiving end, a synthesiser component provided in a vocoder, in which for each of the assigned spectral channels the transmitted parameters control the amplitude of a pulse signal which is influenced in pitch by the fundamental speech frequency or of a noise signal, in dependence upon the tuning character criterion; and wherein the subsidiary speech signals generated in this way are subsequently added to form the synthesised speech signal, the pulse signal responses and the noise signal responses of spectral channel filters are simulated in the receiving end synthesiser component, each being stored in a respective read-only memory, weighted with the respective envelope curve values assigned to the spectral channels for all the spectral channels within one synthesis interval, being read out from one or the other read-only memory in dependence upon the tuning character criterion, and the number of the scan values of a pulse signal response which are to be read-out is effected in dependence upon the fundamental speech frequency when the read-only memory which contains the pulse signal responses is interrogated.

2. A method as claimed in claim 1, in which the transmitted speech-specific parameters are firstly transferred, in the timing of consecutive analysis intervals, into a working store which forwards the envelope curve values to an intermediate store of a multiplier-adder, and forwards to respective control inputs of a control unit the parameters which relate to the fundamental speech frequency and to the tuning character criterion, and that during a synthesis interval, in dependence upon the parameters with which it is supplied at its input, the control unit controls the read-only memories, the multiplier-adder, and the connection of that read-only memory from which data is currently to be read out to the multiplier-adder.

3. A method as claimed in claim 1 or claim 2, in which the transmission of the speech-specific parameters is effected in coded form in a t.d.m. frame determined by the analysis interval, and the speech synthesis is carried out at the receiving end in digital mode, and that at the output of the or a multiplier-adder the digital synthesised speech signal is converted into analogue form by a digital-to-analogue converter which is itself controlled by the control unit.

4. A method as claimed in any preceding claims in which the number of the scan values which are to be read out in respect of a pulse signal response in the course of an interrogation of the read only memory which contains the pulse signal responses, and which relate to a fixed scanning interval, is equal to the ratio of the fundamental period duration to the scanning interval, rounded off to a whole number.

5. A method as claimed in any preceding claim. in which, in the course of the interrogation of the read-only memory which contains the noise signal responses, the number of the scan values which are to be read-out in respect of a noise signal response, and which relate to a fixed scanning interval, is equal to the ratio of the mean synthesis interval to the scanning interval.

6. A method as claimed in any preceding claim in which, in the course of the interrogation of the read-only memory which contains the pulse signal responses, for a prdetermined analysis interval (Tma) the synthesis interval (Tms) satisfies the equation: Tms = Tma + To/r; where To is the fundamental speech frequency period; and r.is the xeduction factor.

7. A method of transmission of speech signals substantially as described with reference to figures 2 to 5.

CLAIMS Amendments to the claims have been filed as follows 1. A method of transmission of speech signals from a transmitting end to a receiving end, wherein at the transmitting end the signal is conducted to the analysis component of a vocoder for the acquisition of speech-specific parameters, composed of the fundamental speech frequency, the tuning character criterion, and the envelope curve values of a specific number of spectral channels, whose respective frequency positions and band-widths are predetermined by a filter bank; wherein these parameters are assembled and transmitted to the receiving end in the timing of consecutive analysis periods; wherein at the receiving end, a synthesiser component is provided in a vocoder to operate in consecutive synthesis periods having individual durations not differing from the duration of an analysis period by more than one fundamental speech period, in which vocoder for each of the assigned spectral channels the transmitted parameters control either the amplitude of a pulse signal influenced in pitch by the fundamental speech frequency or the amplitude of a noise signal, the selection being made in dependence upon the tuning character criterion; and wherein the subsidiary speech signals generated in this way are subsequently added in to form the synthesised speech signal, the pulse signal responses and the noise signal responses of spectral channel filters are simulated in the receiving end synthesiser component, each being stored in a respective read-only' memory, weighted with the respective envelope curve values assigned to the spectral channels for all the spectral channels within one synthesis period, being read out from one or the other read-only memory in dependence upon the tuning character criterion, and the number of the sample values of a pulse signal response which are' to be readout when the read-only memory which contains the pulse signal responses is interrogated is determined by the fundamental speech frequency.

2. A method as claimed in Claim 1, in which the transmitted speech-specific parameters are firstly transferred, in the timing of consecutive analysis periods, into a working store which forwards the envelope curve values to an intermediate store of a multiplier-adder, and forwards to respective control inputs of a control unit the parameters which relate to the fundamental speech frequency and to the tuning character criterion, and that during a synthesis period, in dependence upon the parameters with which it is supplied at its input, the control unit controls the read-only memories, the multiplier-adder, and the connection of that read-only memory from which data is currently to be read out to the multiplieradder.

3. A method as claimed in Claim 1 or Claim 2, in which the transmission of the speech-specific parameters is effected in coded form in a t.d.m.

frame determined by the analysis period, and the speech synthesis is carried out at the receiving end in digital mode, and that at the output of the or a multiplier-adder the digital synthesised speech signal is converted into analogue form by a digital-to-analogue converter which is itself controlled by the control unit.

4. A method as claimed in any preceding Claims in which the number of sample values which are to be read out in respect of a pulse signal response in the course of an interrogation of the readonly memory which contains the pulse signal responses, and which relate to a fixed sampling time, is equal to the ratio of the fundamental speech period to the sampling time, rounded off to a whole number.

5. A method as claimed in any preceding Claim in which, in the course of the interrogation of the read-only memory which contains the noise signal responses, the number of sample values which are to be read-out in respect of a noise signal response, and which relate to a fixed sampling time, is equal to the ratio of the mean synthesis period to the sampling time.

6. A method as claimed in any preceding Claim in which,in the course of the interrogation of the read-only memory which contains the pulse signal responses, for a predetermined analysis period (Tma) the synthesis period (Tms) satisfies the equation : Tms = Tma + where To is the fundamental speech frequency period; and r is the reduction factor.