EP2362681B1 - Method and device for phase-dependent processing of sound signals - Google Patents

Description

The present invention relates to a method and an apparatus for processing sound signals of at least one sound source. The invention is in the field of digital processing of sound signals recorded with a microphone array. In particular, the invention relates to a method and an apparatus for the phase-dependent or phase-sensitive processing of sound signals recorded with a microphone array.

A microphone array is used when two or more spaced microphones are used to record sound signals (multi-microphone technique). This makes it possible to achieve directional sensitivity in digital signal processing. First of all, the classic "shift and add" or "filter and add" methods should be mentioned, in which a microphone signal is shifted in time or filtered relative to the second, before the signals thus manipulated are added together. In this way it is possible to achieve sound cancellation ("destructive interference") for signals arriving from a particular direction. Since the underlying wave geometry is formally identical to the generation of directivity in radio applications when using multiple antennas, this is also referred to as "beam forming", wherein the "beam" of the radio waves is replaced by the direction of attenuation in the multi-microphone technology. The term "beam forming" has prevailed for microphone array applications as a generic name, although here of a "beam" can actually be no question. Misleadingly, the term is used not only for the classical two- or multi-microphone technique just described, but also for more advanced, non-linear array techniques, for which the analogy with antenna technology no longer applies.

In many applications, the classical method misses the actual desired goal. It often does not help to dampen sound signals that arrive from a certain direction. Rather, it is desirable to forward, as far as possible, only those signals originating from one (or more) particular signal source (s), such as those from a desired speaker.

From the EP 1595427 B1 a method for the separation of sound signals is known. According to the method described therein, the angle and the width of the "directional cone" for the desired signals (actually no cone but a rotational hyperboloid) and the attenuation for unwanted signals outside the directional cone can be controlled by means of parameters. The described method calculates a signal-dependent filter function, wherein the spectral filter coefficients are calculated using a predetermined filter function whose argument is the angle of incidence of a spectral signal component. The angle of incidence is determined by means of trigonometric functions or their inverse functions from the phase angle which exists between the two microphone signal components; This calculation is also spectrally resolved, ie separately for each representable frequency. The angle and width of the directional cone as well as the maximum attenuation are parameters of the filter function.

That in the EP 1595427 B1 The disclosed method suffers from several disadvantages. The results achievable with the method correspond only in free field and in the near field the desired goal to separate sound signals of a particular sound source. In addition, a very low tolerance of the components used and in particular the microphones used is required because disturbances in the phases of Mirkofonsignale have a negative effect on the effectiveness of the process. The required tight component tolerances can be realized at least partially by means of suitable production technologies. However, this often involves higher production costs. Near field / free field restrictions are more difficult to circumvent. One speaks of a free field, if the sound wave unhindered arrives at the microphones 10, 11, that is without reflected on the signal path 12 of the sound source 13, attenuated, or otherwise has been changed, as in FIG. 1a is shown. In the near field, in contrast to the far field, where the sound signal arrives as a plane wave, the curvature of the wavefront becomes even clearer. Although this is actually an undesirable departure from the plane wave based geometry considerations of the method, there is usually a significant similarity to the free field in one significant point. Since the signal or sound source 13 is so close, the phase disturbances by reflections o.ä. usually rather low compared to the useful signal. FIG. 1b shows the use of the microphones 10, 11 and the sound source 13 in a narrow space 14, such as a vehicle interior. However, when used in confined spaces, the phase effects are significant because the reflections of the sound waves on particularly smooth surfaces, such as front or side windows, cause the sound waves to propagate on different sound paths 12 and in the vicinity of the microphones the phase relationship between disturb the signals of the two microphones so much that the result of the signal processing according to the above-mentioned method is unsatisfactory.

The phase disturbances due to the reflections, as in FIG. 1b are shown, cause the spectral components of the sound signal of a signal source 13 seemingly from different directions to the microphones 10, 11 meet. FIG. 2 shows in comparison the directions of incidence in free field ( Fig. 2a ) and reflections ( Fig. 2b ). In the free field, all the spectral components of the sound signal 15 _f1 , 15 _f2 ,..., 15 _fn come from the direction of the sound source (in FIG. 2 not shown). According to FIG. 2b meet the spectral components of the sound signal 16 _f1 , 16 _f2 , ..., 16 _fn due to the frequency-dependent reflections each with very different apparent angles of incidence θ _f1 , θ _f2 , ..., θ _fn on the microphones 10, 11, although the sound signal was generated by a sound source 13. A processing of the sound signals in narrower spaces, in which only sound signals from a certain angle of incidence are taken into account, leads to unsatisfactory results, since As a result, certain spectral components of the sound signal are not or only insufficiently processed, resulting in particular in losses in the signal quality.

A further disadvantage of the known method is that the angle of incidence as a spatial angle for each frequency f must first be calculated using trigonometric functions or their inverse functions from the phase angle present between the two microphone signal components. This calculation is expensive, and the u.a. Required trigonometric function Arccosine (arccos) is defined only in the range [-1, 1], so that, if necessary, a corresponding correction function is necessary.

It is therefore an object of the present invention to propose a method and a device for processing sound signals, which avoid the disadvantages of the prior art as possible and in particular to compensate for phase disturbances or effects that are associated with the signals allow. It is a further object of the invention to propose a method and an apparatus for phase-dependent processing of sound signals which allow to compensate for systematic errors in the microphone signals, for example due to component tolerances, and / or calibration of individual components, e.g. to enable the microphones or the entire device.

According to the invention, a method according to claim 1 or a device according to claim 10 is proposed for this purpose. Furthermore, the invention provides a computer program according to claim 12. Advantageous developments of the invention are defined in the respective subclaims.

The method according to the invention for the phase-dependent processing of sound signals of at least one sound source basically comprises the steps of arranging at least two microphones 10, 11 in a respectively predetermined distance d from each other, detecting sound signals with both Microphones and generating associated microphone signals and processing the microphone signals. In a calibration mode, the following steps are carried out: Determining at least one calibration position of a sound source, separately acquiring the sound signals for the calibration position with both microphones and generating calibration microphone signals assigned to the respective microphone for the calibration position, determining the frequency spectra of the assigned calibration microphone signals, and calculating the phase differences φ ₀ (f) of the assigned calibration microphone signals. Since a separate phase difference value is determined for each frequency f, φ ₀ (f) is also referred to below as phase difference vector or frequency-dependent phase difference vector. During an operating mode, the following steps are then performed: acquiring the current sound signals with both microphones and generating associated current microphone signals, determining the current frequency spectra of the associated current microphone signals, calculating a current phase difference vector φ (f) between the assigned current microphone signals from their frequency spectrums, selecting at least one of the determined calibration positions, calculating a spectral filter function F as a function of the current phase difference vector φ (f) and the respective measurement position-specific phase difference vector φ ₀ (f) of the selected calibration position, generating a respective signal spectrum S of a signal to be output by multiplicatively coupling at least one of the two Frequency spectra of the current microphone signals with the spectral filter function F of the respective selected calibration position, wherein the filter function is selected such in that the less the difference between the current and the measurement-position-specific phase difference is for the corresponding frequency, the less the spectral components of sound signals are attenuated, and the respectively output signal for the respective selected calibration position being obtained by inversely transforming the generated signal spectrum.

In this way, the method according to the invention or the device according to the invention provides a calibration procedure, according to US Pat at least one position of the expected useful signal source as a so-called calibration position during the calibration mode sound signals, which are generated for example by playing a test signal, are recorded with their phase effects and - interference from the microphones. From the recorded microphone signals, the phase difference vector φ ₀ (f) between these microphone signals is then calculated from their frequency spectra for the calibration position. In the subsequent signal processing in the operating mode of this phase difference vector φ ₀ (f) is then used to measure the filter function for generating the signal spectrum of the output signal, which can compensate for phase noise and effects in the sound signals. By the subsequent application of the so-filtered filter function on at least one of the current microphone signals by multiplicatively linking the spectrum of the current microphone signal with the filter function, a signal spectrum of the output signal is generated, which contains substantially only signals from the selected calibration position. The filter function is chosen so that spectral components of sound signals that correspond to the Einmessmikrofonsignalen and thus the supposed Nutzsignalen according to their phase difference are not or less attenuated than spectral components of sound signals whose phase difference differs from the einmesspositionsspezifischen phase difference. Furthermore, the filter function is selected such that the greater the difference between the current and calibration position-specific phase difference for the corresponding frequency, the more attenuated spectral components of sound signals are.

If the calibration procedure is applied not only model-specific, but according to one embodiment for each device, such as for each microphone array device in its operating environment, not only the model-typical or environmental, but also by component tolerances and the Operating environment caused to compensate for phase effects and disturbances of the specific device in operation. This embodiment is therefore suitable Component tolerances of the microphones, such as their phase position and sensitivity in a simple and safe way to compensate. In this case, effects that are not caused by changing the spatial position of the useful signal source itself, but by changes in the environment of the useful signal source, for example by opening a side window of a vehicle, can be taken into account. The calibration position is defined as a state spatial position that includes, for example, the state of the room as an additional dimension. If such changes or variations in the calibration position occur during operation, they can not be mastered in principle by a single calibration. For this purpose, the inventive method is then configured as an adaptive method in which the calibration position-specific phase difference vector φ ₀ (f) is calculated or updated not only from microphone signals detected once during the calibration mode, but from the microphone signals of the actual useful signals during operation.

According to one development of the invention, the method or the device initially operates in the operating mode. The calibration position-specific phase difference vector φ ₀ (f) is set to φ ₀ (f) = 0 for all frequencies f. Only at a later time, the method or the device switches to the calibration mode and calculates the calibration position-specific phase difference vector φ ₀ (f), for example, a user speaks test signals and these are detected by the microphones to generate therefrom associated Einmessmikrofonsignale. From the assigned Einmessmikrofonsignalen the einmesspositionsspezifische phase difference vector φ ₀ (f) is then calculated. Subsequently, it is again switched to the operating mode. In which the spectral filter functions F are calculated for each current phase difference vector as a function of the previously determined respective calibration position-specific phase difference vector.

In this way, first use without calibration (often called calibration) under default settings is possible. Once in the Adjustment mode is switched, an adjustment can be achieved not only, for example, in terms of component tolerances but also the current operating environment, the specific conditions of use and the user.

In other words, the invention allows in particular a phase-dependent and at the same time frequency-dependent processing of sound signals, without it being necessary to determine the angle of incidence of the sound signals by at least one spectral component of the current sound signal as a function of the difference between its phase difference and a calibration position-specific phase difference corresponding frequency is attenuated.

• Figur 1 zeigt schematisch die Ausbreitung von Schallsignalen einer Schallquelle im Freifeld (a) und bei Reflexionen im Nahfeld (b).
• Figur 2 zeigt schematisch die scheinbaren Einfallsrichtungen von Schallsignalen einer Schallquelle im Freifeld (a) und bei Reflexionen im Nahfeld (b).
• Figur 3 zeigt ein Ablaufdiagramm zur Bestimmung der Einmessdaten im Einmessmodus gemäß einer Ausführungsform der Erfindung.
• Figur 4 zeigt ein Ablaufdiagramm zur winkelabhängigen Bestimmung der Filterfunktion gemäß einer Ausführungsform der Erfindung.
• Figur 5 zeigt ein Ablaufdiagramm zur phasenwinkelabhängigen Bestimmung der Filterfunktion gemäß einer Ausführungsform der Erfindung.

Brief description of the pictures:

• FIG. 1 schematically shows the propagation of sound signals of a sound source in the free field (a) and reflections in the near field (b).
• FIG. 2 schematically shows the apparent directions of incidence of sound signals of a sound source in the free field (a) and reflections in the near field (b).
• FIG. 3 shows a flowchart for determining the calibration data in the calibration mode according to an embodiment of the invention.
• FIG. 4 shows a flowchart for angle-dependent determination of the filter function according to an embodiment of the invention.
• FIG. 5 shows a flowchart for the phase angle-dependent determination of the filter function according to an embodiment of the invention.

A basic idea of the invention is to determine in a calibration procedure for desired sound signals phase-dependent calibration data which consider the application-related phase effects, and then use these calibration data in signal processing to compensate for phase noise and effects.

The method provides for this an arrangement of at least two microphones 10, 11 at a predetermined distance d to each other. To avoid ambiguity of phase differences, this distance is less than half the wavelength of the highest occurring frequency to choose, ie less than the quotient sound velocity / sampling rate of the microphone signals. A value for the microphone distance d which is well suited in practice for speech processing is, for example, 1 cm. With each microphone, the sound signals, which are generated by a sound source arranged in a measuring position, are then detected separately. Each microphone generates from the sound signals recorded with this microphone Einmessmikrofonsignale assigned to this microphone. From the determined frequency spectra of the assigned Einmessmikrofonsignale a einmesspositionsspezifische phase difference vector φ ₀ (f) is then calculated. The phase differences thus determined between the assigned calibration microphone signals from their frequency spectra are then used in the operating mode as calibration data for the compensation of the corresponding phase disturbances or effects.

zu berechnen.In one embodiment, the calibration data is generated by the sequence of steps as described in the US patent application Ser FIG. 3 are shown flowchart shown. First, in step 310, the playback of a test signal, such as white noise, from the Einmessposition as the position of the expected useful signal source and the recording of the corresponding Einmessmikrofonsignale with the microphones 10 and 11 by separately detecting the sound signals with the two microphones and generating the assigned Einmessmikrofonsignale for this calibration position. Subsequently, the Fourier transforms M1 (f, T) and M2 (f, T) of the calibration microphone signals at time T and the real and imaginary parts Re1, Im1, Re2, Im2 of Fourier transform M1 (f, T) and M2 (f, T) calculated in step 320 to get out of it again at step 330, the frequency-dependent phases φ (f, T) at time T between the calibration microphone signals according to the formula: $$\mathrm{\phi }\left(\mathrm{f},\mathrm{T}\right)=\arctan \left(\left(\mathrm{re}\mathrm{1}*\mathrm{in\; the}\mathrm{2}-\mathrm{in\; the}\mathrm{1}*\mathrm{re}\mathrm{2}\right)/\left(\mathrm{re}\mathrm{1}*\mathrm{re}\mathrm{2}+\mathrm{in\; the}\mathrm{1}*\mathrm{in\; the}\mathrm{2}\right)\right)$$

to calculate.

In a next step 340, the phase vectors φ (f, T) determined at continuous times T are then averaged over T over time, resulting in a calibration position-specific phase difference vector φ ₀ (f) containing the calibration data.

For an angle-dependent filter determination, as described below with reference to FIG. 4 is optionally described in step 350, the calculation of a Einmess angle vector θ ₀ (f) = arccos (φ ₀ (f) c / 2πfd) after correction of the argument to the allowed value range [-1 ... 1]. In an angle-dependent filter determination according to Fig. 4 In contrast to the phase-angle-dependent filter determination, the inverse function of the cosisus (arccos) is required in order to determine a geometric or spatial angle from the phase angle (hence also called partially space-angle-dependent filter determination).

In an angle-dependent filter determination for generating an output signal s (t) in the operating mode according to FIG. 4 First, the current sound signal with the two microphones 10 and 11 is recorded in step 410. In step 420, the Fourier transforms M1 (f, T) and M2 (f, T) of the microphone signals 1 and 2 are again calculated at time T and their real and imaginary parts Re1, Im1, Re2, Im2. Subsequently, in step 430, the frequency-dependent phases at time T φ (f, T) = arctan ((Re1 * Im2-Im1 * Re2) / (Re1 * Re2 + Im1 * Im2)) and from this in turn an angle vector θ (in step 440) f) = arccos (φ (f) c / 2πfd) including corresponding correction of the argument to the allowed value range [-1 ... 1] for all frequencies f. In step 450, the spectral filter function containing the attenuation values for each frequency f at time T and defined as follows is then: F (f, T) = Z (θ (f, T) -θ ₀ (f)) of a unimodal mapping function such as Z (θ) = ((1 + cosθ) / 2) ⁿ where n> 0 is calculated as a function of the calibration angle vector θ ₀ (f), where the angle θ is defined such that -π ≤ θ ≤ π holds. The value n is referred to below as the width parameter, since it defines the adjustable width of the directional cone. It should be noted that the larger the width parameter n is chosen, the smaller the beam width. The thus determined filter function F (f, T) with a value range 0 ≦ F (f, T) ≦ 1 is then applied in step 460 to a spectrum of the microphone signals 1 or 2 in the form of a multiplication: S (f, T) = M1 (FIG. f, T) F (f, T). From the spectrum S (f, T) filtered in this way, the output signal s (t) is then generated in step 470 by inverse Fourier transformation. The above definition of the filter function F (f, T) is to be understood as an example, other assignment functions with similar characteristics fulfill the same purpose. The soft transition chosen here between the extreme values of the filter function (zero and one) has a favorable effect on the quality of the output signal, in particular with regard to unwanted artifacts of the signal processing.

berechnet. Die spektrale Filterfunktion wird nun gemäß der Formel $$\mathrm{F}\left(\mathrm{\varphi }\left(\mathrm{f},\mathrm{T}\right)\right)={\left(\mathrm{1}-{\left(\left(\mathrm{\varphi }\left(\mathrm{f},\mathrm{T}\right)-{\mathrm{\varphi }}_{\mathrm{0}}\left(\mathrm{f}\right)\right)\mathrm{c}/\mathrm{2}\mathrm{\pi fd}\right)}^{\mathrm{2}}\right)}^{\mathrm{n}}\mathrm{mit\; n}>\mathrm{0},$$

im Hinblick auf den einmesspositionsspezifischen Phasendifferenzvektor ϕ₀(f) in Schritt 540 berechnet, wobei c die Schallgeschwindigkeit, f die Frequenz der Schallsignalkomponenten, T die Zeitbasis der Spektrumserzeugung, d der vorbestimmte Abstand der beiden Mikrofone, und n der Breitenparameter für den Richtkegel ist. Beim Betrachten der Formel, welche wie zuvor exemplarisch zu verstehen ist, wird klar, dass die Filterfunktion im Idealfall, d.h. bei Phasengleichheit zwischen aktuell im Betriebsmodus gemessenem und einmesspositionsspezifischem Phasendifferenzvektor, gleich Eins wird, so dass die auf das Signalspektrum S angewandte Filterfunktion das auszugebende Signal nicht dämpft. Bei zunehmender Abweichung des aktuellen vom einmesspositionsspezifischen Phasendifferenzvektor geht die Filterfunktion gegen Null, was zu einer entsprechenden Dämpfung des auszugebenden Signals führt.According to a development of the invention, the determination of the angle is dispensed with and instead, during the calibration procedure, only the measuring-position-specific phase difference vector φ ₀ (f) is determined, which already contains the calibration information. In this embodiment, the calculation of the angle vector θ ₀ (f) in step 350 and thus the possibly necessary correction of the range of values of the argument for the arccos calculation are thus not required in the determination of the calibration data. During the operating mode, the method comprises the in FIG. 5 illustrated steps. First, in turn, the current sound signal with the two microphones 10 and 11 is detected in step 510. From the microphone signals 1 and 2 generated therefrom, the current frequency spectra are calculated by calculating the Fourier transforms M1 (f, T) and M2 (f, T) at time T and their real and imaginary parts Re1, Im1, Re2, Im2 determined in step 520. Subsequently, in step 530, the current phase difference vector is determined from the frequency spectra thereof $$\mathrm{\phi }\left(\mathrm{f},\mathrm{T}\right)=\arctan \left(\left(\mathrm{re}\mathrm{1}*\mathrm{in\; the}\mathrm{2}-\mathrm{in\; the}\mathrm{1}*\mathrm{re}\mathrm{2}\right)/\left(\mathrm{re}\mathrm{1}*\mathrm{re}\mathrm{2}+\mathrm{in\; the}\mathrm{1}*\mathrm{in\; the}\mathrm{2}\right)\right)$$

calculated. The spectral filter function will now be according to the formula $$\mathrm{F}\left(\mathrm{\phi }\left(\mathrm{f},\mathrm{T}\right)\right)={\left(\mathrm{1}-{\left(\left(\mathrm{\phi }\left(\mathrm{f},\mathrm{T}\right)-{\mathrm{\phi }}_{\mathrm{0}}\left(\mathrm{f}\right)\right)\mathrm{c}/\mathrm{2}\mathrm{\pi fd}\right)}^{\mathrm{2}}\right)}^{\mathrm{n}}\mathrm{with\; n}>\mathrm{0}.$$

with respect to the calibration position-specific phase difference vector φ ₀ (f) in step 540, where c is the sound velocity, f is the frequency of the sound signal components, T is the time base of the spectrum generation, d is the predetermined distance of the two microphones, and n is the width parameter for the directional cone. When considering the formula, which is to be understood as an example, it becomes clear that the filter function is equal to one in the ideal case, ie at phase equality between phase difference vector currently measured in the operating mode and measuring position-specific phase difference vector, so that the filter function applied to the signal spectrum S is the signal to be output does not dampen. With increasing deviation of the current measuring position from the specific phase difference vector, the filter function goes to zero, resulting in a corresponding attenuation of the output signal.

If several phase difference vectors have been determined in the calibration mode for, for example, different calibration positions, it is possible to determine the filter function for one of these calibration positions and thus a desired position of the useful signal.

erzeugt, woraus dann wiederum im Schritt 560 das auszugebende Signal s(t) durch inverse Fouriertransformation von S(f,T) bestimmt wird.In step 550, the signal spectrum S of the measured signal is then applied by applying the filter function F (f, T) to one of the microphone spectra M1 or M2 in the form of a multiplication according to the formula (here for microphone spectrum M1): $$\mathrm{S}\left(\mathrm{f},\mathrm{T}\right)=\mathrm{M}\mathrm{1}\left(\mathrm{f},\mathrm{T}\right)\mathrm{F}\left(\mathrm{f},\mathrm{T}\right)$$

is generated, from which in turn, in step 560, the signal s (t) to be output is determined by inverse Fourier transformation of S (f, T).

According to a development of the invention, the method initially operates in the operating mode and the calibration position-specific phase difference vector φ ₀ (f) is set equal to zero for all frequencies f to φ ₀ (f). This corresponds to a so-called "Broadview" geometry without calibration. If the device for processing sound signals is now to be measured, the device is switched to the calibration mode. Assuming now that a corresponding useful signal is generated, for example by only the desired user speaking, the calibration position-specific phase difference vector φ ₀ (f) is calculated. In this case, the user speaks, for example, predetermined test sets, which are detected by the microphones and from which associated Einmessmikrofonsignale be generated. For example, the system or device enters the calibration mode by an external command in which it determines the φ ₀ (f). For this purpose, the user speaks test sounds, eg "sch sch sch", until the system has collected sufficient calibration data, which can optionally be displayed, for example, by an LED. The system then changes to the operating mode in which the calibration data is used.

The system then switches to the operating mode and the spectral filter function F is calculated for each current phase difference vector as a function of the previously determined respective calibration position-specific phase difference vector. Thus, it is possible, for example, to first deliver the device, such as a mobile phone, in a basic setting and then the calibration procedure with the voice of the actual user in the user's preferred environment and arrangement, ie how does the user hold the mobile in relation to the user's mouth User or similar, perform.

According to a development of the invention, the width parameter n is selected to be smaller in the operating mode with the previously calculated respective calibration position-specific phase difference vector than in the initially assumed operating mode than in the unacceptable operating state in which the device is in a default setting. An initially smaller width parameter means a wider beam, so that initially tend to be less attenuated sound signals from a larger beam direction. Only when the calibration has taken place, the width parameter is chosen to be larger, because now the filter function is able to properly attenuate the sound signals arriving at the microphones, taking into account the (phase) noise occurring in the near field according to a smaller directional cone. The beamwidth defined by the parameter n in the mapping function is e.g. selected smaller in operation with calibration data than in unadjusted case. By measuring the method knows the position of the signal source so very accurate, so that you can then work with a "sharper" beam forming and therefore with a narrower beam than in the unreasonable case, where the position of the source is at most approximately known.

According to one embodiment of the invention, in the calibration mode, the calibration position is further varied in a spatial and / or state region in which the user is expected in the operating mode. Subsequently, the calibration position-specific phase difference vector φ ₀ (f) is calculated for these varied calibration positions. In addition to different spatial positions, other effects which are caused, for example, by an opened side window of a motor vehicle, can then be taken into account during calibration, since not only the position of the user, for example the seat position of the driver of the motor vehicle, but also the surrounding state, ie For example, the side window is open or closed, take into account.

In principle, fluctuations occurring during operation can not be controlled by a single calibration. This comes in accordance with a According to a further development of the invention, an adaptive method is used which evaluates the actual useful signals during operation instead of calibration signals. According to such an embodiment, an adaptive post-measurement is carried out only in such a situation in which, apart from the useful signal, no other noise signals are picked up by the microphones, which can be recognized, for example, by the relative constancy of the phase difference vectors φ (f, T) at successive times T. is.

According to one embodiment, the method is designed as an adaptive method. The calibration position-specific phase difference vector φ ₀ (f) is initially set equal to zero either for all frequencies f at φ ₀ (f) or, for example, stored values are used for all frequencies of the calibration position-specific phase difference vector φ ₀ (f) from previous calibration or operating modes. Alternatively, after initially passing through the calibration mode for calculating the current calibration-position-specific phase difference vector φ ₀ (f), it is switched to the operation mode. In further operation, the calibration position-specific phase difference vector φ ₀ (f) is then updated by the adaptive method by interpreting the current sound signals of a sound source in the operating mode as sound signals of the selected calibration position and used for the calibration. Thus, an update of the calibration data unnoticed by the user is used, wherein the update always takes place when it is assumed that the current sound signals are noise-affected useful signals in the sense of the respective application or the current configuration of the device, so that from these Sound signals then the einmesspositionsspezifische phase difference vector φ ₀ (f) is determined. An otherwise possibly predetermined by the device switching between calibration and operating mode can thus be omitted. Rather, the measurement takes place "subliminal" during operation whenever it allows the signal quality. A criterion for the signal quality can be, for example, the signal-to-noise ratio of the microphone signals.

However, the effect on the signal to be output of a disc lowered during operation can continue to be compensated inadequately or not at all in this manner, since the boundary condition of the freedom from interference noise during the detection of the sound signals for determining the calibration data can hardly be realized in this case. In order to make the adaptation noise-resistant, according to an embodiment of the invention, therefore, a constantly updated, spectrally resolved noise estimate is performed, the estimated noise being subtracted from the microphone spectra before the adaptation process, before the actual compensation of the phase effects is performed. According to one embodiment, the method therefore further comprises that interference signals are first calculated out of the microphone signals of the current sound signals in the operating mode with the aid of a tracking, phase-sensitive noise model before the calibration-position-specific phase difference vector φ ₀ (f) is updated.

According to a development of the invention, the step of specifying at least one calibration position further comprises arranging a test signal source in or near the calibration position, transmitting a calibrated test signal through the sound signal source, acquiring the test signal with the two microphones, and generating the associated one Einmessmikrofonsignale alone from the test signal. So far, it has been assumed that the phase angle φ _{0 is} spectrally resolved, ie frequency-dependent, and the corresponding vector φ ₀ (f) is determined during the calibration procedure on the basis of the recorded test signals, whereas the broad-determining parameter n is scalar, ie the same for all frequencies. Defining a half-value phase difference φ _½ (f) at which the filter function F (φ (f, T)) has dropped to the value 1/2, the width parameter n depends on φ _½ (f) in the above definition of the filter function F (FIG. φ (f, T)) is composed as follows: $$\mathrm{n}=-\mathrm{1}/{\mathrm{<figure-callout\; id="2"\; label="log"\; filenames="imgf0004.png,imgf0005.png"\; state="\{\{state\}\}">log</figure-callout>}}_{\mathrm{2}}\left(\mathrm{1}-{\left({\mathrm{c\phi }}_{\mathrm{½}}\left(\mathrm{f}\right)/\mathrm{2}\mathrm{<figure-callout\; id="2"\; label="\pi fd"\; filenames="imgf0004.png,imgf0005.png"\; state="\{\{state\}\}">\pi fd</figure-callout>}\right)}^{\mathrm{2}}\right).$$

φ _½ (f) is a parameter vector, which is initially given for each frequency f.

For an extended calibration procedure, the source of the test signals, for example a so-called artificial mouth, is no longer positioned only at the location of the expected useful signal source, but varies over a spatial range in which a variation of the position of the useful signal source is to be expected during normal operation. In a motor vehicle application, for example, this is intended to cover the fluctuation range which is caused by natural head movements, variable seat adjustments and different body sizes of a driver. For each measurement with different locations of the test signal source, a vector φ ₀ (f) is now determined as described above. Subsequently, the arithmetic mean values μ (f) and the standard deviations σ (f) for each frequency f are calculated from the measurements for each frequency over the calculated calibration-position-specific phase difference vectors φ ₀ (f). It should be noted that the mean values μ (f) are arithmetic mean values of previously time-averaged variables; μ (f) is now used instead of φ ₀ (f). The previously scalar parameter n is now also made frequency-dependent and determined by the calibration procedure. For this purpose, the half-value phase difference φ _½ (f) is linked to the standard deviation via a constant k φ _½ (f) = k σ (f). If a normal distribution is now assumed for the measured values φ ₀ (f), which is not necessarily the case, but for better knowledge according to the method is nevertheless assumed, would be 95% of all measurement results within the range ± φ _½ (f), if k = 2 selects. For the broad determining parameter n (f) then: $$\mathrm{n}\left(\mathrm{f}\right)=-\mathrm{1}/{\mathrm{<figure-callout\; id="2"\; label="log"\; filenames="imgf0004.png,imgf0005.png"\; state="\{\{state\}\}">log</figure-callout>}}_{\mathrm{2}}\left(\mathrm{1}-{\left(\mathrm{c\sigma }\left(\mathrm{f}\right)/\mathrm{<figure-callout\; id="2"\; label="\pi fd"\; filenames="imgf0004.png,imgf0005.png"\; state="\{\{state\}\}">\pi fd</figure-callout>}\right)}^{\mathrm{2}}\right)\mathrm{,}$$

With this extension of the Einmessvorgangs one takes into account the fact that not only the incidence or phase angle by reflections Frequency-dependent, but that the strength of this change can be frequency-dependent, which can be compensated by a spectrally resolved "beam width" according to the method.

Furthermore, it should be mentioned that all described devices, methods and process components are of course not limited to use, for example in a motor vehicle. In the same way, e.g. Also, a mobile phone or any other (voice) signal processing device are used that uses a microphone array technology.

The method according to the invention and the device according to the invention are expediently implementable by means of or in the form of a signal processing system, for example with a digital signal processor (DSP system) or as a software component of a computer program running, for example, on a PC or DSP system or any other hardware platform.

LIST OF REFERENCE NUMBERS

10

11 spaced microphones;

M1 (f, T), M2 (f, T)

Fourier transforming the microphone signals (spectral amplitude at frequency f at time T);

d

Distance between microphones MIK1 and MIK2;

f

Frequency;

T

Time of determination of a spectrum or an output signal

φ ₀ (f)

time averaged phase difference vector in the calibration mode;

φ (f, T)

Phase difference vector of the microphone signals during operation;

Re1 (f), Im1 (f)

Real and imaginary parts of the spectral components of the first hands-free microphone signal (microphone 1);

Re2 (f),

Im2 (f) real and imaginary parts of the spectral components of the second handsfree microphone signal (microphone 2);

θ ₀ (f)

time averaged frequency-dependent angle of incidence of the first test audio signal in the calibration mode;

θ (f, T)

frequency-dependent angle of incidence of the microphone signals during operation;

μ (f)

arithmetic mean values for each frequency f over the φ ₀ (f);

σ (f)

Standard deviations for each frequency f over the φ ₀ (f);

n

Width parameter;

n (f)

frequency-dependent width parameter, where φ _½ (f) = kσ (f), where φ _½ (f) is the frequency-dependent phase difference at which the filter function F assumes the value 1/2 at the frequency f;

F (f, T)

Filter function;

Z

unimodal assignment function;

S (f, T)

Signal spectrum of the signal to be output;

s (t)

output signal.

Claims (12)

1. A method for phase-sensitive processing of sound signals of at least one sound source, comprising the steps of:

   - arranging two microphones (10, 11) at a distance d from each other;

   - capturing sound signals with both microphones, and generating associated microphone signals; and

   - processing the sound signals of the microphones;

   wherein during a calibration mode, the method comprises the following steps:

   - defining at least one calibration position of the sound source where the sound source is positioned in the calibration mode;

   - capturing separately the sound signals for the at least one predetermined calibration position with both microphones, and generating associated calibration microphone signals for the calibration position;

   - determining the frequency spectra of the associated calibration microphone signals;

   - calculating a calibration-position-specific phase difference vector ϕ₀(f) between the associated calibration microphone signals from their frequency spectra for the at least one predetermined calibration position;

   the method further comprising the following steps during an operating mode in which a desired signal source is positioned at or near one of the at least one predetermined positions:

   - capturing the current sound signals with both microphones and generating associated current microphone signals;

   - determining the current frequency spectra of the associated current microphone signals;

   - calculating a current phase difference vector ϕ(f) between the associated current microphone signals from their frequency spectra at the time T;

   - selecting one of the at least one predetermined calibration position as a position of the desired signal source;

   - calculating a spectral filter function F depending on the current phase difference vector and the respective calibration-position-specific phase difference vector of the selected calibration position;

   - generating, respectively, a signal spectrum S of a signal to be output by multiplication of at least one of the two frequency spectra of the current microphone signals with the spectral filter function F of the respective selected calibration position, the filter function being chosen so that the smaller the absolute value of the difference between current and calibration-position-specific phase difference for the corresponding frequency, the smaller the attenuation of spectral components of sound signals; and

   - obtaining the respective signal to be output for the respective selected calibration position by inverse Fourier transformation of the generated signal spectrum.
2. The method according to claim 1, the method further comprising the following steps during calibration mode:

   - calculating M1(f, T) and M2(f,T) of the spectral amplitude at frequencies f at time T by Fourier transformation of the calibration microphone signals;

   - calculating the real and imaginary parts Re1, Im1, Re2, Im2 of the Fourier transforms M1(f, T) and M2 (f, T);

   - calculating the phase difference ϕ(f, T) at time T between the calibration microphone signals, according to the formula: $$\mathrm{\varphi }\left(\mathrm{f},\mathrm{T}\right)=\arctan \left(\left(\mathrm{Re}\mathrm{1}*\mathrm{Im}\mathrm{2}-\mathrm{Im}\mathrm{1}*\mathrm{Re}\mathrm{2}\right)/\left(\mathrm{Re}\mathrm{1}*\mathrm{Re}\mathrm{2}+\mathrm{Im}\mathrm{1}*\mathrm{Im}\mathrm{2}\right)\right);$$

   

   
   and

   - averaging ϕ(f, T) temporally over subsequent times T in order to obtain the calibration-position-specific phase differences ϕ₀(f), which contain the result of the calibration process.
3. The method according to claim 2, the method during operating mode further comprising the following steps:

   - calculating the spectral filter function according to the formula: $$\mathrm{F}\left(\mathrm{\varphi }\left(\mathrm{f},\mathrm{T}\right)\right)={\left(\mathrm{1}-{\left(\left(\mathrm{\varphi }\left(\mathrm{f},\mathrm{T}\right)-{\mathrm{\varphi }}_{\mathrm{0}}\left(\mathrm{f}\right)\right)\mathrm{c}/\mathrm{2}\mathrm{\pi fd}\right)}^{\mathrm{2}}\right)}^{\mathrm{n}},$$

   

   
   where n > 0;

   - generating the signal spectrum S by applying the filter function F(f,T) to a microphone spectrum M1 in the form of a multiplication, according to the formula: $$\mathrm{S}\left(\mathrm{f},\mathrm{T}\right)=\mathrm{M}\mathrm{1}\left(\mathrm{f},\mathrm{T}\right)\mathrm{F}\left(\mathrm{f},\mathrm{T}\right);$$

   

   - generating the output signal s(t) by inverse Fourier transformation of S(f,T);

   where:

   c is the speed of sound,

   f is the frequency of the sound signal components,

   T is the time base of the spectrum generation,

   d is the distance between the two microphones, and

   n is a parameter n>0 to be determined, whereby the width of the filter function is determined.
4. The method according to any one of the preceding claims, the method first working in operating mode, and the calibration-position-specific phase difference vector ϕ₀ (f) being set to ϕ₀(f) = 0 for all frequencies f, and the method further comprising:

   - switching into calibration mode and calculating the calibration-position-specific phase difference vector ϕ₀(f), a user speaking test signals, which are captured by the microphones, and associated calibration microphone signals are generated from them;

   - switching into operating mode, and calculating the spectral filter function F for each current frequency-dependent phase difference vector depending on the respective previously determined calibration-position-specific phase difference vector.
5. The method according to claim 4, related to claim 3, wherein in operating mode with the previously calculated calibration-position-specific phase difference vector, the width parameter n is chosen to be greater than in the initially taken uncalibrated operating mode.
6. The method according to any one of the preceding claims, the method in calibration mode further comprising the following steps:

   - varying the calibration position in a spatial and/or state range in which the user is expected in operating mode;

   - calculating calibration-position-specific phase difference vectors ϕ₀(f) for varied calibration positions;

   - calculating the arithmetic means µ(f) and standard deviations σ(f) for each frequency f of the calculated calibration-position-specific phase difference vectors ϕ₀(f); and

   the method during operating mode further comprising the following steps:

   - calculating the spectral filter function according to the formula: $$\mathrm{F}\left(\mathrm{\varphi }\left(\mathrm{f},\mathrm{T}\right)\right)={\left(\mathrm{1}-{\left(\left(\mathrm{\varphi }\left(\mathrm{f},\mathrm{T}\right)-{\mathrm{\varphi }}_{\mathrm{0}}\left(\mathrm{f}\right)\right)\mathrm{c}\right)}^{\mathrm{2}}\right)}^{\mathrm{n}\left(\mathrm{f}\right)}$$

   

   
   with a frequency-dependent width parameter n(f), according to the formula: $$\mathrm{n}\left(\mathrm{f}\right)=-\mathrm{1}/{\log }_{\mathrm{2}}\left(\mathrm{1}-{\left(\mathrm{c\sigma }\left(\mathrm{f}\right)/\mathrm{\pi fd}\right)}^{\mathrm{2}}\right);$$

   

   - generating the signal spectrum S by applying the filter function F(f,T) to a microphone spectrum M1 in the form of a multiplication according to the formula: $$\mathrm{S}\left(\mathrm{f},\mathrm{T}\right)=\mathrm{M}\mathrm{1}\left(\mathrm{f},\mathrm{T}\right)\mathrm{F}\left(\mathrm{f},\mathrm{T}\right);$$

   

   where:

   c is the speed of sound,

   f is the frequency of the sound signal components,

   T is the time base of the spectrum generation,

   d is the distance between the two microphones,

   n(f) is the frequency-dependent width parameter which is defined by ϕ_1/2(f) = kσ(f), and

   ϕ_1/2(f) is the frequency-dependent phase difference at which the filter function F at frequency f takes the value F(f) = 1/2.
7. The method according to any one of the preceding claims, wherein the step of defining at least one calibration position further includes:

   - arranging a test signal source near the specified calibration position;

   - the sound signal source sending a calibrated test signal;

   - both microphones capturing the test signal, and the associated calibration microphone signals being generated from the test signal only.
8. The method according to one of claims 1 to 6, the method being in the form of an adaptive method, and wherein after passing through calibration mode initially, a switch into operating mode takes place to calculate the current calibration-position-specific phase difference vector ϕ₀(f), and in further operation, the calibration-position-specific, frequency-dependent phase difference vector ϕ₀(f) is updated, the current sound signals of a sound source being interpreted in operating mode as sound signals of the selected calibration position.
9. The method according to claim 8, wherein interference signals are first calculated out of the microphone signals of the current sound signals in operating mode using a concurrent, phase-sensitive noise model, before the calibration-position-specific phase difference vector ϕ₀(f) is updated.
10. A device for phase-sensitive processing of sound signals of at least one sound source, comprising:

    - two microphones (10, 11), which are arranged at a predetermined distance (d) from each other, to capture sound signals and generate microphone signals;

    - a processing unit which is connected to the microphone unit, to process the microphone signals;

    wherein during a calibration mode, the processing unit with the microphones is set up and at least one calibration position of the sound source is determined, wherein the processing unit further is adapted for:

    - capturing separately the sound signals for the calibration position with both microphones, and generating associated calibration microphone signals for the at least one determined calibration position;

    - determining the frequency spectra of the associated calibration microphone signals;

    - calculating a calibration-position-specific phase difference vector ϕ₀(f) between the associated calibration microphone signals from their frequency spectra for the at least one determined calibration position; and

    during an operating mode, the processing unit with the microphones is adapted and one of the at least one determined calibration positions is selected as a position of a desired signal source, and the processing unit further is adapted for:

    - capturing the current sound signals with both microphones and generating associated current microphone signals;

    - determining the current frequency spectra of the associated current microphone signals;

    - calculating a current phase difference vector ϕ(f) between the associated current microphone signals from their frequency spectra at the time T;

    - calculating a spectral filter function (F) depending on the current phase difference vector and the respective calibration-position-specific, frequency-dependent phase difference vector of the selected calibration position;

    - generating, respectively, a signal spectrum S of an output signal by multiplication of at least one of the two frequency spectra of the current microphone signals with the spectral filter function F of the respective selected calibration position, the filter function being chosen so that the smaller the absolute value of the difference between current and calibration-position-specific phase difference for the corresponding frequency, the smaller the attenuation of spectral components of sound signals; and

    the device further includes an output unit to output the signal to be output for the relevant selected calibration position, with means for inverse Fourier transformation of the respective generated signal spectrum.
11. The device according to claim 10, further set up to carry out the method according to one of claims 2 to 9.
12. A computer program containing program code which, when executed on a data processing unit, realizes the method of one of claims 1 to 9.

Images