Method for the automatic switching of the speech direction and circuit arrangement for implementing the method.

The invention relates to a method for automatic speech switching in which a received signal with variable attenuation is fed to a loudspeaker, in which a signal emitted by a microphone with variable attenuation is provided as a transmit signal, in which the received signal and microphone signal are each continuously classified as a speech signal or noise , in which the attenuation of the one signal that has been classified as a speech signal is set to a first attenuation value and the respective other signal is set to an attenuation value that is greater than the first, in which, in the case of classification of both signals as speech signals, the previous attenuation values ​​are retained, in which, in the case of classification of both signals as noise, both attenuations to one between first and second damping value lying third damping value can be set.

Furthermore, the invention relates to a circuit arrangement for carrying out this method with a controllable receive attenuator, at the input of which the received signal is applied and at the output of which there is a signal intended to control the loudspeaker, a controllable transmit attenuator, at the input of which the microphone signal is applied and at the output the transmission signal is present, two signal-noise detectors, one of which receives the received signal or the loudspeaker signal and the other the microphone signal or the transmitted signal, and continuously classifies the received signal and transmitted signal as a speech signal or noise and with a control logic downstream of the signal-noise detectors for controlling the reception and Transmitter attenuator, which in one case attenuates the signal branch whose signal has been classified as a speech signal, to a first damping value and the respective other signal (Tx, Rx) to a second damping value that is greater than the first, maintains the previous damping values ​​in the case of classification of both signals as speech signals and one between the first and second in the case of classification of both signals as noise Damping value sets third damping value.

The convenience of telephone sets has increased significantly in recent years. In addition to redialing, storing phone numbers and open listening, the ability to speak freely, ie to make calls without a handset, has become a popular function of the telephone. In addition to use in the private sector, this function can be used in many useful ways, for example in the field of office communication, such as in conference calls, when used in a car phone or in all other cases in which holding the handset is a hindrance.

The basic difference between a normal telephone with a handset and a hands-free telephone is the mode of operation: the former works in intercom mode, ie there is a transmission in both directions - i.e. sending and receiving - simultaneously. With the hands-free telephone, this is only possible with very complex and unsatisfactory procedures. Because of the high signal amplification in both directions, any attempt to make conversation in intercom mode would immediately lead to a strong feedback whistle, since the acoustic coupling between loudspeaker and microphone creates a closed loop. Handsfree telephones can therefore only work in intercom mode, ie only one of the two subscribers can speak at a time while the other is listening. To achieve this, you need a circuit that detects who is currently speaking, then switches through the relevant channel and attenuates the other sufficiently. The loop gain is thus kept below one. If the speakers swap their functions, the circuit must determine this immediately and switch the channels accordingly. With the help of voice message recognition in conjunction with an electronic switch, the circuit then works automatically in hands-free mode.

A method and a circuit arrangement for automatic switching of voice messages in telephones is known, for example, from European patent application 0 423 537.

In this method or in this circuit arrangement, the currently inactive channel is damped compared to the active channel. This means that the speaker who is speaking cannot hear his counterpart. In addition, a so-called idle state can also be controlled in which both channels are operated with half the maximum damping. In the event that both are silent or if the speaker is drowned out by the noise at the location of the listener, the system will slowly switch to idle state (slow idle). In the event that the speaker and the loudspeaker have approximately the same level, on the other hand, the system quickly changes to the idle state (fast idle). It is only possible to change the direction of speech if the person who has been speaking is silent and the person who has been listening speaks.

It is therefore not possible to interrupt the speaker. If an interference signal occurs in the reception branch which is incorrectly classified as a speech signal, it is therefore not possible to change the direction of speech at all.

The object of the invention is therefore to provide a method for automatic switching of speech messages and a circuit arrangement for carrying out the method which does not have this disadvantage.

This object is achieved in a method of the type mentioned in that the signal supplied to the loudspeaker and the microphone signal are compared with one another, and the transmitted signal and the received signal are compared with one another in terms of a variable which can be assigned to the respective signal volume, in that with an attenuation set to the second attenuation value Microphone signal in the case of a volume that is louder than the loudspeaker signal by a certain first difference, recognized as a speech signal microphone signal, the attenuation of the microphone signal is set to the first attenuation value and the attenuation of the received signal to the second attenuation value and otherwise the previous attenuations are retained and that with an attenuation of the received signal set to the second attenuation value in the case of a by a certain second difference louder than the broadcast signal, received signal recognized as speech signal, the attenuation of the received signal is set to the first attenuation value and the attenuation of the microphone signal is set to the second attenuation value, and otherwise the previous attenuations are retained.

In the circuit arrangement of the type mentioned at the outset, the object is achieved in that a comparison device which compares the signal fed to the loudspeaker and the microphone signal with one another and the transmitted signal and the received signal with one another with respect to a variable which can be assigned to the respective signal volume, is coupled to the control logic and that the control logic at a transmit attenuator set to the second attenuation value in the case of a microphone signal louder by a certain amount than the loudspeaker signal, recognized as a speech signal, the attenuation of the transmit attenuator is equal to the first attenuation value and the attenuation of the receive attenuator equal to the second damping value, in the case of a reception attenuator set to the second attenuation value, in the case of a reception signal which is louder by a certain amount compared to the transmission signal and recognized as a speech signal, sets the reception attenuator to the first attenuation value and the attenuation of the transmission attenuator to the second attenuation value and otherwise maintains the previous attenuations.

Further developments and refinements of the inventive concept are characterized in the subclaims.

The invention is explained in more detail below with reference to the exemplary embodiments shown in the figures of the drawing. Show it:<dl id="dl0001"><dt>Figure 1</dt><dd>the basic process sequence in a preferred circuit arrangement for carrying out the method in a block diagram,</dd><dt>Figure 2</dt><dd>the damping curve generated by the method according to the invention in a diagram,</dd><dt>Figure 3</dt><dd>the preferred procedure for distinguishing speech and noise in a circuit arrangement for performing the method of Figure 1 in a block diagram,</dd><dt>Figure 4</dt><dd>a preferred embodiment of a low pass used in a block diagram.</dd></dl>

In the exemplary embodiment according to FIG. 1, signals transmitted bidirectionally via a telephone line are split into a reception signal Rx and a transmission signal Tx in a circulator device ZE. The receive and transmit signals Rx, Tx are fed or taken from an automatic speech direction switch device DS according to the invention. Furthermore, a loudspeaker LS, to which a signal Lx is fed, and a microphone MIC, which emits a signal Mx, are connected to the speech direction switching device DS via a loudspeaker adapter LA. The speech direction switch device DS has a controllable receive attenuator AR, at the input of which the received signal Rx is applied and at the output of which the signal Lx is present, and a controllable transmit attenuator AT, at whose input the signal Mx is applied and at the output of which the transmit signal Tx is applied . A control logic ACL is provided to control the reception and transmission attenuator AR, AT, which in turn is controlled by two signal-noise detectors NMR, NMT. One signal-noise detector NMR has the received signal Rx and the other signal-noise detector NMT has the signal Mx applied to it. In principle, however, it is also possible to use the signal-noise detectors NMR; NMT with the signal Lx or to control the transmit signal Tx, since the signal Lx is proportional to the receive signal Rx and the transmit signal Tx is proportional to the signal Mx.

In addition, the control logic ACL is coupled to a comparison device CC, which on the one hand compares the signals Lx and Mx with one another and the signals Tx and Rx with one another with respect to a variable that can be assigned to the respective signal volume. The individual signal pairs can be compared with one another continuously and next to one another, a signal pair then still having to be selected according to certain criteria, or only the signal pair relevant for the respective operating state is compared. In the embodiment shown in Figure 1, the latter option is realized. The comparison circuit CC contains a comparator KP, the output signal of which is fed to the control logic ACL and the inputs of which are preceded by a peak value detector PD1 and PD2. Both peak value detectors PD1 and PD2 have a discharge time constant of, for example, 512 msec or work with a fixed decrement value, since the peak value detectors PD1 and PD2 operate strictly logarithmically due to preceding logarithmic amplifiers LA1 and LA2. A logarithmic amplifier LA1 and LA2 is connected upstream of the peak value detectors, each of which ensures compression of the signal fed to them and thus reduces the required dynamic range of the peak value detectors PD1 and PD2. In addition, the compression results in a rating that corresponds to the perception of volume. The logarithmic amplifiers LA1 and LA2 in turn are each a level adjustment stage GS1 or GS2 connected upstream, which generates a constant damping or amplification value that is expected on the basis of the acoustic echo or fork echo. Each level adjustment stage GS1 or GS2 can in turn be divided into two amplifier stages. A first stage GS2a could emulate the echo path and takes into account a possible signal clipping due to the finite modulation range. Such a signal clipping can be caused by a possible level increase / amplification between the signal Lx fed to the loudspeaker and the signal Mx emitted by the microphone. The second stage GS2b would implement a lead necessary due to the signal runtime. Like GS2, GS1 could be implemented in two amplifier stages. Depending on the operating case, whether the signals Lx and Mx or the signals Tx and Rx are compared with one another, the amplifier stages GS1 and GS2 and the peak value detectors PD1 and PD2 can be set differently. The control logic ACL switches over to the values ​​assigned to the operating case. Alternatively, the amplifier stages GS1 or GS2 could also be connected in each case between the logarithmic amplifier LA1 or LA2 and peak value detector PD1 or PD2. Either the received signal Rx and the transmitted signal Tx or the signal Lx fed to the loudspeaker and the signal Mx emitted by the microphone are applied to the level matching stages GS1 and GS2 by means of a switching device SU. In addition, the switching device SU applies an offset signal to an input of the comparator KP, depending on the operating case. The offset signal 01 is fed to either an adder AD1 or an adder AD2 which, depending on the operating case, superimposes the offset signal 01 on the output signal of the phase detector PD1 or PD2. The switching unit SU is controlled by the control unit ACL, which in turn evaluates the output signal of the comparator KP.

The decrements of the peak value detectors PD1 and PD2, as well as the amplification elements GS1 and GS2 could be implemented in a switchable manner for each operating case. This would result in two working conditions for each operating case. On the acoustic side, on which the signals Lx and Mx are compared, it is possible to switch depending on the received signal Rx classified as speech or noise. This would allow a distinction to be made between the following system behaviors: short-term, but large increases in level during speech due to resonances and direct coupling are optimally controlled with a larger GS1 and larger decrements or smaller time constant, while in the case of only noise after a hold time with a second parameter set GS1 and PD1 only the level increase due to the room echo and the indirect coupling is provided. On the line side, on which the signals Tx and Rx are compared, different echo properties can be taken into account depending on the input signal Mx classified as speech or noise. So z. B. the near echo plus the far echo can be checked in speech and only the near echo of your own fork can be compensated for its level in the case of noise.

During operation of the arrangement shown in FIG. 1, it is provided that the received signal Rx with variable attenuation is supplied to the loudspeaker LS as signal Lx and that the signal Mx is also used with variable attenuation as transmit signal Tx. The respective damping (or amplification) is set as a function of the received signal Rx and the signal Mx. For this purpose, both signals are continuously classified as speech signal or noise. In the event that one of the two signals has been classified as a speech signal, the attenuation of this signal is set to a first attenuation value D1 and the other signal to a second attenuation value D2 and is maintained until the one signal is again classified as noise . If both signals are classified as a speech signal, the attenuations are set according to the signal that was first recognized as speech. If both signals are classified as noise, both attenuations are set to a third attenuation value between the first and second attenuation values. The transition from the first or second damping value D1, D2 to the third damping value D3 is slower than a transition from the third damping value D3 to the first or second damping value D1, D2 or as a transition from the first damping value D1 to the second damping value D2 and vice versa. This situation is shown graphically in FIG. 2. There, the signal Mx is initially classified as a speech signal and the received signal Rx as a noise. The attenuation of the signal Rx is therefore equal to the second attenuation value D2 and the attenuation of the signal Mx is equal to D1, where D1 is equal to the attenuation zero. The attenuation is adjusted accordingly with the receive attenuator for the received signal Rx and with the transmit attenuator for the signal Mx. At a time T1 the signal Mx changes so that both signal-noise detectors NMR, NMT now classify the corresponding signals Rx, Mx as noise. The transition of the damping from the value D2 to the value D3 in the received signal Rx and the transition of the damping from the value D1 to the value D3 in the signal Mx takes place continuously until a time T3, where both damping values ​​remain until a time T4. At time T4, the signal Mx is recognized again as a speech signal. The damping of the signal Mx is then brought to the damping value D1 with a steep edge and the received signal Rx to the value D2. At time T4, the direction of speech changes, so that the attenuations also reverse accordingly. According to the invention, the person who is listening now has the option of changing the language direction by drowning out the speaker (ie also an interference signal recognized as language) by speaking louder in contrast. For this purpose, the comparison device CC, depending on the respective operating state, compares either the signal Lx supplied to the loudspeaker and the microphone signal Mx with one another, or the transmission signal Tx and the reception signal Rx with one another with respect to a variable that can be assigned to the respective signal volume. The control logic ACL then sets the attenuation of the transmit attenuator AT equal to the first attenuation value and the attenuation of the receive attenuator AR in the case of a transmission attenuator AT set to the second attenuation value AT in the case of a microphone signal Mx which is louder by a certain amount than the loudspeaker signal Lx and recognized as a speech signal the second damping value, while in the case of a reception attenuator AR set to the second attenuation value, in the case of a reception signal Rx which is louder by a certain amount compared to the transmission signal Tx and recognized as a speech signal, it sets the reception attenuator AR to the first attenuation value and the attenuation of the transmission attenuator AT to the second attenuation value. This means that depending on the direction of speech one of the signal pairs, namely receive and transmit signal Rx, Tx on the one hand and loudspeaker signal Lx and microphone signal Mx on the other hand, are examined and with a striking difference, which in the exemplary embodiment is caused by the offset signal 01 and the amplification / attenuation the level adjustment level GS1 or GS2 is given, leads to a change in the direction of speech. In all other cases, the damping is maintained, which is determined by the control logic ACL using the signal-noise detectors NMR, NMT. The control unit SU is accordingly controlled by the control logic ACL. Instead of a single evaluation unit CC that switches between the two signal pairs, two separate comparison devices can also be provided, the control logic ACL only taking into account the comparison result that is of interest in the respective operating case. In addition, a further circuit variant is provided in that instead of an offset signal two different offset signals, which are applied to one input of the comparator KP, or a differential offset signal is provided which, depending on the operating case, inverts or not inverts to both inputs is activated. Furthermore, instead of peak value detection, it is also possible, for example, to measure the power of the individual signals and compare them for an assessment of the signal volume. An embodiment with only one comparison circuit, which in the embodiment of the invention, the level adjustment stages GS1 and GS2 are adapted to the respective attenuation conditions in accordance with the respective speaking direction. During operation of the arrangement shown in FIG. 1, the received signal Rx with variable attenuation is supplied as the signal Lx to the loudspeaker LS and the signal Mx is also used with variable attenuation as the transmit signal Tx. The respective attenuation is set as a function of the received signal Rx and the signal Mx. For this purpose, both signals are continuously classified as voice signals or noise and compared with one another in terms of volume. In the event that one of the two signals has been classified as a speech signal, the attenuation of this signal is set to a first attenuation value D1 and the respective other signal to a second attenuation value D2.

In an embodiment of the invention, the transition of the attenuations caused by the signal change at the time T1 takes place after a given delay time only at the time T2 when the received signal Rx and the signal Mx are classified as noise at time T1. The duration of the transition can be maintained or, as shown, can also run in a shorter time.

To classify the received signal Rx and signal Mx as a speech signal or noise, the magnitude of the respective signal Rx, Mx is first formed and an amplitude companding is then carried out. After a first low-pass filtering with a given time constant, a peak value detection with two alternative decay time constants and a second low-pass filtering with two alternative time constants, a comparison of the signal after the peak value detection minus an offset signal with the signal after the second low-pass filtering is carried out. The classification of the signal as a speech signal and, associated with this, an adjustment of the larger time constant in the peak value detection and the second low-pass filtering results from the predominance of the signal after the peak value detection minus the offset signal compared to the signal after the second low-pass filtering. In the opposite case, the signal is classified as noise and the smaller time constant is set for peak value detection and second low-pass filtering. It is also possible to dispense with peak value detection and time constant switching in the second low-pass filtering for numerous applications.

The classification of speech signal and noise is based on the fact that speech signals generally have a temporally strongly structured envelope with pronounced amplitude fluctuations and can therefore be viewed approximately as a pulse-shaped signal, whereas noises are predominantly relatively uniform, stationary signals. In order to distinguish impulsiveness from uniformity, the signal to be examined is passed on the one hand via a low-pass filtered branch (second low-pass filtering) and on the other hand via a direct branch in which only an offset signal, preferably a direct signal, is subtracted from the signal . The low-pass filtering in one branch attenuates pulse-shaped signals more than the uniform ones. The second branch is independent of the signal type. Thus, when comparing the two branches, a distinction can be made in that the first branch delivers a lower value for pulse-like signals than the second and vice versa for uniform signals. The preceding low-pass filtering, with short interference pulses being suppressed by the low-pass filtering (first low-pass filtering), makes the structure of the signals more distinctive and easier to distinguish. The amplitude companding increases the sensitivity with a small modulation and thus achieves a faster response of the signal-noise detector. A further improvement is achieved in a further development of the invention by the peak value detection and the switching of the time constants in the case of peak value detection and second low-pass filtering, by using a smaller time constant in the case of uniform signals than in the case of pulse-shaped signals, as a result of which a change from noise to speech signal is recognized immediately, while at conversely, the observation time is longer, so that smaller pauses do not lead to a switch.

For this purpose, according to FIG. 3, in the present exemplary embodiment, a rectifier AV is provided at the input E as signal-noise detector NMR, NMT, followed by a compander CP, which in turn is followed by a first low-pass filter LP1, followed by a peak value detector PD with an adjustable time constant. An evaluation circuit CU at output A, to which the output signal of the peak value detector PD minus the offset signal 02 and minus the output signal of the second low pass LP2 is fed, controls the time constants of the peak value detector PD and the second low pass LP2 via an output.

Developing the invention further, the time constant of the first low-pass filter LP1 in the signal-noise detector NMT for the signal Mx is greater than that of the first low-pass filter LP1 in the signal-noise detector NMR for the received signal Rx. This has the advantage that the disruptive influence of echoes occurring on the transmission path between loudspeaker LS and microphone MIC is reduced.

In addition, in the signal Mx or in the reception signal Rx, the signal before the second low-pass filtering is set to a higher value than the instantaneous amplitude value when the classification of the other signal, i.e. the reception signal Rx or the signal Mx, changes from speech signal to noise . A preferred value is the maximum amplitude value that can be represented. As a result, a longer delay time for suppressing interfering echoes is advantageously achieved.

In a further development of the invention, it is provided that the offset signal 02 can be set separately for both signal-noise detectors NMR, NMT. In the exemplary embodiment shown, the offset signal 02 of the one signal-noise detector is set larger for a certain time according to the invention when the other signal-noise detector has classified its received signal as a speech signal. Interfering echoes are suppressed even more effectively. It is also advantageous to suppress the interference signal 02 and / or the instantaneous amplitude value of the other signal-noise detector for a certain time in order to suppress interference after the transition from speech to noise in one signal-noise detector.

The method according to the invention is preferably carried out by time and amplitude-discrete signal processing. The advantages here are precise reproducibility without parameter variation, greater integrability with little external wiring, and greater compatibility with digital communication networks, such as ISDN systems (ISDN = Integrated Services Digital Network). In the exemplary embodiment shown, the transmit and receive signals Tx, Rx and the signals Mx, Lx are digital. The microphone matching device MA and the loudspeaker matching device are thus provided, among other things, for analog-digital or digital-analog conversion.

In a system that is discrete in time and amplitude, low-pass filtering is carried out, for example, as follows: The digital signal to be filtered is multiplied by a constant factor a and added to the filtered digital signal and delayed by a time value that is equal to the reciprocal of the work cycle. From this, the filtered signal multiplied by the factor a is subtracted, which represents the output signal. The factor a is preferably chosen equal to an nth power of 2, so that multiplications can be replaced by shifting operations by n places that are much easier to implement.

A circuit implementation is shown in FIG. 4. The signal IS to be filtered is fed to a first shift register SR1 and shifted arithmetically to the right by n places. The output of the shift register SR1 is fed to a summer SUM, which is also connected to the output of a delay element VE and, via a sign reverser -1, to the output of a second shift register SR2. The input of the second shift register SR2 is connected to the output of the delay element VE and is provided for arithmetic shifting to the right by n positions of this signal. The delay element VE is driven by the output of the summer SUM, which carries the output signal OS. The time constant TC of such a filter stage is calculated as a function of the number of digits n shifted and a time value DT that is equal to the reciprocal of the work cycle, as follows:<maths id="math0001" num=""><math display="inline"><mrow><mtext>TC = DT / {- 1n (1 - a)} with a = 2 ''</mtext></mrow></math><img file="EP0642251A2_D0001.tif" /></maths> According to FIG. 2, a transition from the damping value D2 to the value D3 takes place, for example, by continuous subtraction of the instantaneous value of the damping arithmetically shifted by k digits from the instantaneous value until the smaller damping value is reached, and a transition from the damping value D1 to the value D3 by continuous addition of the instantaneous value arithmetically shifted to the right by k digits until the greater damping value D2 or D3 is reached. Here, too, it is advantageous that no multiplications are required. In addition, however, other realizations are also possible in principle.

The method according to the invention ensures that the loop gain is less than one in every operating case, in particular when both speakers are active. Compared to known circuits, the method according to the invention for implementing them requires a considerably lower circuit complexity than known ones. Due to the slow and continuous transition to an intermediate value of attenuation between the two extreme values, a more pleasant hearing impression is generated when changing from a speech signal to a noise. This is further improved by an additional delay time between signal change and attenuation change when there is no intercom and by measures that prevent switching by echoes.