(19)
(11)EP 1 633 121 A1

(12)EUROPEAN PATENT APPLICATION

(43)Date of publication:
08.03.2006 Bulletin 2006/10

(21)Application number: 04021055.1

(22)Date of filing:  03.09.2004
(51)Int. Cl.: 
H04M 9/08  (2006.01)
G10L 21/02  (2006.01)
H04R 3/00  (2006.01)
(84)Designated Contracting States:
AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HU IE IT LI LU MC NL PL PT RO SE SI SK TR
Designated Extension States:
AL HR LT LV MK

(71)Applicant: Harman Becker Automotive Systems GmbH
76307 Karlsbad (DE)

(72)Inventors:
  • Kellermann, Walter, Prof.Dr.-Ing
    91058 Erlangen (DE)
  • Herbordt, Wolfgang,c/oKellermann, W., Prof.Dr.-Ing
    91058 Erlangen (DE)

(74)Representative: Grünecker, Kinkeldey, Stockmair & Schwanhäusser Anwaltssozietät 
Maximilianstrasse 58
80538 München
80538 München (DE)

  


(54)Speech signal processing with combined adaptive noise reduction and adaptive echo compensation


(57) The present invention relates to a system for speech signal processing with combined noise reduction and echo compensation, comprising an adaptive beamforming signal processing means (Fig.1,w(k);Fig.2 wc(k),B(k) and wa(k)) configured to receive first speech signals (x(k),M) and to derive adaptively beamformed signals from the first speech signals, an adaptive echo compensating means (Fig.1,Fig.2,a(k)) configured to reduce acoustic echoes in the adaptively beamformed signals and an adaptation means configured to simultaneously perform a combined adaptation of the adaptive beamforming means and the adaptive echo compensating means.




Description

Description of the Invention



[0001] The present invention relates to a system and a method for multi-channel speech signal processing with combined noise reduction and acoustic echo compensation.

[0002] Speech signal processing has often to be performed in a noisy background environment. A prominent example is hands-free voice communication in vehicles. Hands-free telephones provide a comfortable and safe communication systems of particular use in motor vehicles. In the case of hands-free telephones it is mandatory to suppress noise in order to guarantee the communication. The amplitudes and frequencies of the noise signals are temporally variable due to, for example, the speed of the vehicle and road noises.

[0003] Of particular importance is the suppression of signals of the remote subscriber which are emitted by the loudspeakers and therefore received again by the microphone(s), since otherwise unpleasant echoes can severely affect the quality and intelligibility of voice conversation. In the worst case acoustic feedback can even leads to a complete breakdown of communication.

[0004] One main problem lies in the reverberating characteristics of the room. In the case of hand-free telephones in motor vehicles the detection of the interior acoustics becomes very difficult, since abrupt changes in the acoustics, for example, owing to movements of the vehicle occupants have to be distinguished.

[0005] To overcome the above mentioned problems means for acoustic echo compensation (AEC) are provided, that basically work at follows. By means of a linear (EP 0204718 B1) or non-linear (EP 1020068 B1; WO-A-96/26592) adaptive filter a replica of the acoustic feedback is synthesized and a compensation signal is obtained from the received signal of the loudspeakers. This compensation signal is subtracted from the sending signal of the microphone thereby generating a resulting signal to be sent to the remote subscriber.

[0006] Besides the acoustic echoes, hands-free communication suffers from a relatively low signal-to-noise ratio (SNR). Consequently, some noise reduction must be employed in order to improve the speech signal quality.

[0007] Current multi-channel systems primarily make use of adaptive or non-adaptive beamformers, see, e.g., "Optimum Array Processing, Part IV of Detection, Estimation, and Modulation Theory" by H.L. van Trees, Wiley & Sons, New York 2002. The beamformer combines multiple microphone input signals to one beamformed signal with an enhanced SNR.

[0008] The combination of AEC means and adaptive beamformers is known for the art (see, e.g., "Microphone Arrays: Signal Processing Techniques and Applications", by Kellermann, W., Springer, Berlin, 2001). However, depending on the actual implementations different convergence properties of the adaptive beamformers and the echo canceling means can inhibit synergetic effect and/or the necessity of employing one echo compensating means for each microphone channel results in uneconomical realizations.

[0009] Despite the recent developments and improvements, effective echo compensation and noise reduction in speech signal processing proves still to be a major challenge. It is therefore the problem underlying the present invention to overcome the above-mentioned drawbacks and to provide a system and a method for speech signal processing, in particular, suitable for hands-free telecommunication system showing improved acoustic properties as an enhanced signal-to-noise ratio (SNR) and efficient acoustic echo compensation.

[0010] The problem is solved by a system for speech signal processing with noise reduction and echo compensation according to Claim 1 and a method according to Claim 12. The system according to the present invention comprises an adaptive beamforming signal processing means configured to receive first speech signals and to derive adaptively beamformed signals from the first speech signals, an adaptive echo compensating means configured to reduce acoustic echoes in the adaptively beamformed signals and an adaptation means configured to simultaneously perform a combined adaptation of the adaptive beamforming means and the adaptive echo compensating means.

[0011] The inventive system may further comprise a microphone array comprising at least two microphones to detect microphone signals as well as at least one loudspeaker to output loudspeaker signals comprising second speech signals, and wherein the first speech signals mentioned above may be obtained on the basis of the microphone signals, and the adaptive echo compensating means may reduce acoustic echoes in the adaptively beamformed signals on the basis of the loudspeaker signals.

[0012] Such a system for speech signal processing provides a significantly enhanced SNR and, ideally, a complete echo canceling. By speech signals any signals representing information of verbal utterances are meant. The speech signals processed by the inventive system, e.g., might be available as functions of discrete time indices or as Fourier transformed signal in the frequency domain. The microphones not only detect speech signals but also noise and acoustic echoes. Consequently, the adaptive beamforming means not only receives speech signals but also noise and echo signals.

[0013] The beamforming and the echo compensating means are adaptive ones which means that they employ filters comprising temporal varying filter vectors that are continuously re-calculated during the speech communication in order to optimize the speech signal processing with respect to noise reduction and echo compensation. The echo compensating means can be a single- or multiple-channel one.

[0014] The simultaneous and combined adaptation of the adaptive beamforming means and the adaptive echo canceling means gives rise to desirable synergetic effects.

[0015] Technically speaking the combined adaptation is realized by the adaptation of an effective weight vector representing the combined weights of the adaptive filters of both the beamforming means and the AEC means.

[0016] Different from a typically state-of-the-art system that is characterized by first employing an adaptive beamforming means and subsequently employing an AEC means, wherein the AEC means suffers from a low convergence speed and thus, usually cannot contribute significantly to the echo canceling, both noise reduction and echo canceling are performed very efficiently by the inventive system. Contrary to the art, the inventive combined adaptation of the adaptive beamforming and the adaptive AEC means allows to simultaneously update both means, i.e. no tracking problems occur at all.

[0017] One should also note that only one echo compensating means is necessary according to the present invention. This is different from constructions know from the art wherein first AEC means and subsequently an adaptive beamforming means are employed requiring one AEC means for each microphone channel.

[0018] Moreover, the number of degrees of freedom for both the noise reduction and the echo canceling are increased by the number of loudspeakers thereby improving the optimization of the adaptive filters employed by the beamforming and AEC means, respectively.

[0019] According to one embodiment of the present invention the adaptive beamforming signal processing means is an adaptive beamformer.

[0020] This realization of the inventive system seems to be the simplest and cheapest one. A beamformer combines different input signals received concurrently to one single beamformed signal. The noise levels of the signal portions in each channel usually differ from each other. By simply averaging the different channels with very low SNR may deteriorate the SNR of the output signal. Thus, it is desirable to introduce weighting coefficients (weights) for the different channels, that are preferably calculated time-dependently in response to changing speech detecting condition, e.g., time-dependent noises and/or speaker movement. According to one advantageous embodiment the beamformed signal is obtained by summing up the weighted concurrently received input signals.

[0021] Depending on the actual application an improved performance may, however, be desired.

[0022] Thus, according to another inventive embodiment the adaptive beamforming signal processing means comprises a blocking matrix configured to receive the first speech signals and to obtain noise reference signals, an adaptive noise canceling means configured to receive the noise reference signals and to obtain adaptive noise signals, a beamformer configured to receive the first speech signals and to obtain first beamformed signals, and a means configured to subtract the adaptive noise signals obtained by the adaptive noise canceling means from the first beamformed signals to obtain the adaptively beamformed signals.

[0023] A generic structure comprising the beamformer, the blocking matrix, which can be chosen as being a non-adaptive or an adaptive one, and the adaptive noise canceling means is a General Sidelobe Canceller (GSC). GSC designs comprise two signal processing path: one noise reduction and one reference signal processing path. The reference signal processing path comprises of a beamformer which is directed to one or more sources of desired signals detected by the microphone array. The microphone signals detected concurrently by the microphone array and comprising noise and echoes besides the desired speech signal are combined to one beamformed signal with an enhanced SNR.

[0024] The noise reduction signal processing path usually comprises a blocking matrix receiving the speech signals and employed to generate noise reference signals. In the simplest realization the blocking matrix performs a subtraction of adjacent channels of the received signals.

[0025] The noise reference signals may serve as input signals for an adaptive noise canceling means. The adaptive noise cancellation is done by (complex-valued) filters. These filters are usually adapted with the goal that the power of the output signal is minimized. In a conventional GSC the output signal of the adaptive noise canceling means is subtracted form the beamformed signal to achieve an output signal. According to the present invention, the adaptive noise cancellation is combined with an adaptive echo canceling. The filters of the combined adaptive noise canceling means and the adaptive echo canceling means are adapted simultaneously.

[0026] The incorporation of a GSC structure significantly reduces the processor load. Different form previous combinations of a GSC and an AEC means, the inventive simultaneous and combined adaptation of GSC and AEC components inhibits leakage of acoustic echoes through the blocking matrix and the adaptive noise canceling means, even when acoustic echoes are efficiently suppressed.

[0027] Also, different from the art the inventive system works reliably even when strong local noise signals are detected by the microphone array and/or the echo paths are changing continuously.

[0028] The adaptive noise canceling means and the adaptive echo compensating means comprise adaptive filters and preferably the filter length of the adaptive filter of the adaptive echo compensating means is equal to or longer than the filter length of the adaptive filter of the adaptive noise canceling means.

[0029] By "filter length" commonly the number of filter coefficients is meant. Usually, the AEC means needs a higher number of filter coefficients to work sufficiently reliable which results in a lower convergence speed as compared to the adaptive beamforming means. In noisy environments with relatively low reverberations, however, e.g. in vehicle, the choice of a comparable or equal number of filter coefficients for the AEC means and the beamforming means may be tolerable.

[0030] It may be desirable to have the adaptation means to simultaneously perform the combined adaptation of the adaptive beamforming means and the adaptive echo compensating means by applying a Least Square optimization criterion.

[0031] The Least Square optimization represents a well-tested and robust method in speech signal processing for adapting complex filters and it is relatively simple to employ. It is performed by minimizing a windowed sum of squared output signal samples subject to constraints that assure at least uniqueness of the output signal. The windowing function employed in the summation selects samples of output signals to be considered for the optimization procedure.

[0032] The constraints of the Least Square optimization criterion may be formulated as


where the upper index T denotes the vector transposition operation, C is the conventional constraint matrix known from linearly-constrained minimum variance beamforming, w(k) and a(k) are the filters of the adaptive beamforming and the adaptive noise canceling, respectively, Q is the number of the at least one loudspeaker, Na is the number of filter coefficients of the filters of the adaptive echo canceling means and c is the constraint vector representing C constraints.

[0033] These constraints represent the combined adaptation of the filters of the adaptive beamforming means and the adaptive echo compensating means by means of the matrix (wT (k),aT (k))T. The dimension of the null matrix C x QNa results from the reasonable assumption that the loudspeaker signals are uncorrelated.

[0034] The system according to the present invention may further comprise a means configured to perform a time delay compensation of the microphone signals to obtain the first speech signals which subsequently are used as input signals for the adaptive beamforming means be it the adaptive beamformer or a GSC like structure.

[0035] It may be advantageous to synchronize the microphone signals corresponding to a desired target signal in order to further improve performance of the inventive system. For this purpose, the time delay of each signal has to be computed and compensated.

[0036] It may be preferred to further employ an adaptive self-calibration means configured to match the phases and amplitudes of the microphone signals or the time delay compensated microphone signals to obtain the first speech signals.

[0037] The adaptive filters of the adaptive self-calibration means can be calculated by means of the Normalized Least-Mean Square algorithm. After time delay compensation the microphone signals may not be matched accurately due to, e.g., speaker movement and phase and amplitude mismatch of the different microphones. By the adaptive self-calibration, the mismatches with respect to phases and amplitudes are minimized. Accordingly, the desired signals in each of the channels are time (phase)-aligned, the amplitudes of the desired signal portions are almost equal in each of the channels and the signals are expected to exhibit very similar frequency characteristics.

[0038] The adaptive beamforming signal processing means and the adaptive echo compensating means and/or the means configured to perform a time delay compensation of the microphone signals and/or the adaptive self-calibration means are configured to perform processing in the time domain or in the frequency domain or in the sub-band frequency domain.

[0039] A realization of the inventive system working a the frequency domain may be considered as being advantageous, since adaptation the frequency domain may show better convergence properties and better tracking capabilities. Suitable means for performing Fast Fourier Transforms and Inverse Fast Fourier Transforms where it is appropriate are known from the art.

[0040] The present invention also provides a system for speech signal processing as described above, wherein the microphone array comprises at least one directional microphone.

[0041] Moreover, the present invention provides a hands-free system comprising the inventive system for speech signal processing with combined noise reduction and echo compensation as described above.

[0042] Furthermore, the present invention is directed to the use of the previously described systems for speech signal processing and hands-free systems in a vehicle. An improved SNR and echo canceling is particularly desirable in the acoustically challenging contents of communication devices in vehicular cabins. If the microphone array is arranged in a frame, it can easily be mounted in a vehicular cabin.

[0043] Also, it is provided by the present invention a vehicle comprising one of the above-described systems for speech signal processing and hands-free systems.

[0044] Furthermore, the present invention provides a method for speech signal processing with combined noise reduction and echo compensation, comprising receiving first speech signals as input signals for an adaptive beamforming employing adaptive beamforming filters, receiving input signals for an adaptive compensating of acoustic echoes employing adaptive echo compensation filters, determining combined adaptive filters from the adaptive beamforming filters and the adaptive echo compensation filters, simultaneously adapting the adaptive beamforming filters and the adaptive echo compensation filters by adapting the combined adaptive filters and obtaining adaptively beamformed signals by the adaptive beamforming and echo compensating the adaptively beamformed signals to obtain an output signal of the speech signal processing.

[0045] The adaptive beamforming as well as the adaptive echo compensating are performed utilizing respective adaptive filters. The associated weight vectors can be combined to one combined filter vector. The adaptation of this combined filter vector results in a simultaneous adaptation of the filters of the combined adaptive beamforming means and the adaptive echo compensating means.

[0046] The inventive method preferably comprises detecting microphone signals by a microphone array comprising at least two microphones and obtaining loudspeaker signals comprising second speech signals that are output by at least one loudspeaker. In this case the first speech signals mentioned above can be obtained on the basis of the microphone signals, and the loudspeaker signals can be used as the input signals for the adaptive compensating of acoustic echoes.

[0047] The adaptive beamforming may be performed by an adaptive beamformer or by the steps of receiving and processing of the first speech signals by a blocking matrix to obtain noise reference signals, processing of the noise reference signals by an adaptive noise canceling means to obtain adaptive noise signals, receiving and processing of the first speech signals by a beamformer to obtain first beamformed signals and subtracting the adaptive noise signals obtained by the adaptive noise canceling means from the first beamformed signals to obtain adaptively beamformed signals.

[0048] Adaptive filters are used for the adaptive echo compensation and the adaptive noise canceling and the filter length of the filter of the adaptive echo compensating means is preferably equal to or longer than the filter length of the filter of the adaptive noise canceling means.

[0049] The combined adaptive filters are preferably adapted by applying a Least Square optimization criterion with a constraint according to


where the upper index T denotes the vector transposition, C is the conventional constraint matrix known from LCMV beamforming, w(k) and a(k) are the filters of the adaptive beamforming and the adaptive noise canceling, respectively, Q is the number of the at least one loudspeaker, Na is the number of filter coefficients of the filters of the adaptive echo canceling means and c is the constraint vector representing C constraints.

[0050] The microphone signals may advantageously be compensated for time delay to obtain the first speech signals and the microphone signals or the time delay compensated microphone signals may be adaptively self-calibrated.

[0051] The adaptive beamforming and the adaptive echo compensating and/or the time delay compensating and/or the adaptive self-calibrating can be performed in the time domain or in the frequency domain or in the sub-band frequency domain.

[0052] Furthermore, the present invention provides a computer program product, comprising one or more computer readable media having computer-executable instructions for performing the steps of the inventive method for speech signal processing with noise reduction as described above.

[0053] Additional features and advantages of the present invention will be described with reference to the drawings. In the description, reference is made to the accompanying figures that are meant to illustrate preferred embodiments of the invention. It is understood that such embodiments do not represent the full scope of the invention.

[0054] Figure 1 shows one embodiment of the speech signal processing means according to the present invention comprising an adaptive beamformer and an AEC means being simultaneously optimized.

[0055] Figure 2 shows another embodiment of the speech signal processing means according to the present invention comprising a GSC and an AEC means being simultaneously optimized.

[0056] Figure 3 shows selected steps of the inventive method for speech signal processing with combined noise reduction and acoustic echo compensation.

[0057] The general structure of one embodiment of the speech signal processing according to the present invention is illustrated in Figure 1. For illustration purposes, the processing is done in the discrete time domain. Alternatively, the algorithms could be applied in the full Discrete Fourier Transform domain or the sub-band domain.

[0058] The microphone signals detected by M microphones 11 can be available as x(k) (k is the discrete time index) and are assumed to be superimposed signals


where d(k), n(k) and e(k) denote the desired, the local noise and the acoustic echo signals, respectively. The echoes are caused by the output signals of at least one loudspeaker 12 and show the reverberating characteristics of the room.

[0059] The signals x(k) are processed by a combined adaptive beamformer. The beamformer is designed to combine the input signals of the different M microphone channels to one single beamformed output signal with an enhanced signal-to-noise ratio (SNR). For this purpose the beamformer makes use of adaptive filters w(k). The AEC means models the echo paths, e.g., as linear responses. By means of adaptive filters a(k) the echo portions of the processed microphone signals can be subtracted.

[0060] The output signal of the combined system can be written as a function of the sensor signals x(k), the loudspeaker signals v(k) as well as the adaptive filters w(k) and a(k):

with




where the upper index T denotes the vector transposition operation. The loudspeaker signals v(k) of Q loudspeaker channels are, in general, the reference signals for the acoustic echoes, i.e. they serve as input signals for the adaptive echo compensating means 14. The signals v(k) can be regarded as further sensor channels or paths of the beamformer thereby increasing the degrees of freedom usable for the adaptive beamforming.

[0061] The adaptive beamformer, i.e. the adaptive filters w(k) with Nw filter coefficients, and the AEC means, i.e. the adaptive filters a(k) with Na filter coefficients, are now optimized in combination and simultaneously by an adaptation means (in Figure 1 it is indicated by the dashed box), in a way that the beamformer gain is maintained and the acoustic echoes are suppressed. For the optimization procedure the Least Square (LS) optimization criterion can be applied. In principle, the mathematical technique is similar to the one applied, e.g., to conventional Linear Constrained Minimum Variance (LCMV) beamformers (see, e.g., "Optimum Array Processing, Part IV of Detection, Estimation, and Modulation Theory" by H.L. van Trees, Wiley & Sons, New York 2002). One has to solve for

with


where C (k) is the conventional constraint matrix (of size M Nw x C) known from LCMV beamformers and c(k) is the constraint column vector of length C that represents C constraints to assure unity of the beamformed output signal. The C x QNa nulls in the constraint matrix result from the reasonable assumption that the loudspeaker signal v(k) are uncorrelated. The windowing function wi extracts the desired samples from the output signals which should be included into the optimization procedure.

[0062] It should be noted that according to the optimization criterion above, for an arbitrary number of microphones, only one error signal has to be minimized. The mathematical solution of the optimization problem is state of the art of signal processing.

[0063] Figure 2 illustrates an embodiment of the speech signal processing means according to the present invention comprising a General Sidelobe Canceller (GSC) and an AEC means being simultaneously optimized. The beamformer of Figure 1 is replaced by a GSC structure. The GSC can be regarded as one advantageous realization of an adaptive beamformer and comprises two signal processing paths.

[0064] The reference (or upper) signal processing path comprises of a fixed beamformer 23 directed to one or more sources of a desired signal detected by the microphones 21. The beamformed signal yWc(k) is characterized by an enhanced SNR.

[0065] The noise reduction (or lower) signal processing path comprises of a non-adaptive or an adaptive blocking matrix 25 that generates noise reference signals yB(k). As little desired speech as possible should be present in these signals. In the simplest realization the blocking matrix performs a subtraction of adjacent channels.

[0066] The noise reference signals yB(k) are used as input signals for an adaptive noise canceling means 26. The adaptive noise cancellation is done by (complex-valued) filters wa(k). These filters are usually adapted with the goal that the power of the output signal is minimized. The output signal of the adaptive noise canceling means is then subtracted from the beamformed signal. Since the reference signals ideally contain no signal portions, the residual noise of the signal ywc(k) is almost completely reduced by the adaptive noise canceling means and thereby the SNR of the final output signal y(k) is increased.

[0067] In conventional settings for adaptation of the filters for noise reduction, the Normalized Least-Mean Square (NLMS) algorithm may be used which reads:




where YGSC denotes the output signal of the GSC alone, i.e., if no AEC means was present, β is some positive real-valued number and the asterisk denotes the complex conjugate of the noise reference signals. Accordingly, the noise reference signals and the adaptive filters wa(k) are used to generate an adaptive noise signal to be subtracted from the beamformed signal yWc output by the beamformer 23.

[0068] However, according to present invention the adaptive filters of both the GSC and the AEC means have to be adapted simultaneously by optimizing for w̃(k). The algebra is consequently more involved.

[0069] The weight vector w̃(k), introduced in the description of Figure 1, can be projected onto two orthogonal subspaces, one of which (c(k)) can be chosen to fulfill the C constraints (constraint subspace):



[0070] The second subspace can be chosen as

with

and


where

is the identity matrix of size QNa x QNa and the columns of the matrix B(k) are orthogonal to the columns of the constraint matrix C(k). The least square minimization of the final output signal y(k) now reads

with c(k)being the projection of the weight vector w̃(k) onto the second of the orthogonal subspaces fulfilling the constraint of the optimization problem.

[0071] By the solution of the minimization problem one obtains the optimal weight vector a,opt(k)

with



[0072] By

the sample correlation matrix of the data vector x̃(k) is denoted and the upper index + means the pseudoinverse of the matrix. Formally the solution for the optimum weight vector is equivalent to the optimum weight vector for the GSC (see, e.g., "Broad band beamforming and the generalized sidelobe canceller", IEEE Trans. on Acoustic Speech and Signal Processing, vol. 34, p.1322).

[0073] Figure 3 illustrates basic steps of the inventive speech signal processing according to one preferred embodiment. An array of microphones comprising at least two microphones detects signals 30 comprising of speech signals 31, noise signals 32 and echo signals 33 and produces microphone outputs 34.

[0074] According to this embodiment the output microphone signals are synchronized corresponding to a desired target signal. For this purpose the time delay of each signal has to be computed and compensated 35.

[0075] The synchronized signals serve as input signals for both signal processing paths of the employed GSC. A conventional delay-and-sum beamformer of the upper signal processing path may combine the signals to obtain an output signal 36 with enhanced SNR. A blocking matrix generates noise reference signals 37 by subtracting adjacent channels. Alternatively, a more elaborated matrix known from the art as, e.g., a Walsh-Hadamard kind of blocking matrix or a Griffiths-Jim blocking matrix may be employed.

[0076] The noise reference signals are used as input signals for adaptive noise canceling 38. One practical problem lies in the optimization of the adaptive filters of the noise canceling means. Besides noise reduction the inventive speech signal systems provides echo compensation. An AEC means makes use of the loudspeaker signals 39 to compensate for the acoustical echoes 40.

[0077] According to the present invention the adaptation of the filters of both the GSC and the AEC means is performed simultaneously. The beamformed signal obtained by the beamforming 36 is further processed by the adaptive noise canceling 38 and the adaptive echo compensating 40, in order to achieve an output signal 41 of high purity.

[0078] It may be advantageous to supplement the step of time delay compensation 35 of the microphone output signals by adaptively self-calibrating (not shown) the time delayed signals. Ideally, after this further step of processing the calibrated signals are phase-aligned with respect to the desired signal and the amplitudes of the desired signal portions are equal in each of the multiple channels.

[0079] The complex self-calibration filters that perform a matching of the time delayed signals in each channel, can be adapted by the NLMS algorithm (see above) with the goal that the powers of the error signals are minimized.

[0080] All previously discussed embodiments are not intended as limitations but serve as examples illustrating features and advantages of the invention. It is to be understood that some or all of the above described features can also be combined in different ways.


Claims

1. System for speech signal processing with combined noise reduction and echo compensation, comprising
an adaptive beamforming signal processing means (13) configured to receive first speech signals and to derive adaptively beamformed signals from the first speech signals,
an adaptive echo compensating means (14, 24) configured to reduce acoustic echoes in the adaptively beamformed signals, and
an adaptation means configured to simultaneously perform a combined adaptation of the adaptive beamforming means and the adaptive echo compensating means.
 
2. System according to Claim 1 further comprising
a microphone array (11, 21) comprising at least two microphones to detect microphone signals,
at least one loudspeaker (12, 22) to output loudspeaker signals comprising second speech signals, and
wherein the first speech signals are obtained on the basis of the microphone signals, and
the adaptive echo compensating means (14, 24) reduces acoustic echoes in the adaptively beamformed signals on the basis of the loudspeaker signals.
 
3. System according to Claim 1 or 2, wherein the adaptive beamforming signal processing means is an adaptive beamformer (13).
 
4. System according to Claim 1 or 2, wherein the adaptive beamforming signal processing means comprises
a blocking matrix (25) configured to receive the first speech signals and to obtain noise reference signals,
an adaptive noise canceling means (26) configured to receive the noise reference signals and to obtain adaptive noise signals, and
a beamformer (23) configured to receive the first speech signals and to obtain first beamformed signals, and
a means configured to subtract the adaptive noise signals obtained by the adaptive noise canceling means (26) from the first beamformed signals to obtain the adaptively beamformed signals.
 
5. System according to Claim 4, wherein the adaptive noise canceling means (26) and the adaptive echo compensating means (24) comprise adaptive filters and the filter length of the adaptive filter of the adaptive echo compensating means is equal to or longer than the filter length of the adaptive filter of the adaptive noise canceling means.
 
6. System according to one of the preceding claims, wherein the adaptation means simultaneously performs the combined adaptation of the adaptive beamforming means and the adaptive echo compensating means by applying a Least Square optimization criterion.
 
7. System according to Claim 6, wherein the constraint of the Least Square optimization criterion reads


where the upper index T denotes the vector transposition operation, C is the conventional constraint matrix known from linearly-constrained minimum variance beamforming, w(k) and a(k) are the filters of the adaptive beamforming and the adaptive noise canceling, respectively, Q is the number of the at least one loudspeaker, Na is the number of filter coefficients of the filters of the adaptive echo canceling means and c is the constraint vector representing C constraints.
 
8. System according to one of the Claims 1 - 7, further comprising a means configured to perform a time delay compensation of the microphone signals to obtain the first speech signals.
 
9. System according to one of the Claims 1- 7, further comprising an adaptive self-calibration means configured to match the phases and amplitudes of the microphone signals or the time delay compensated microphone signals to obtain the first speech signals.
 
10. System according to one of the preceding claims, wherein the adaptive beamforming signal processing means and the adaptive echo compensating means and/or the means configured to perform a time delay compensation of the microphone signals and/or the adaptive self-calibration means are configured to perform processing in the time domain or in the frequency domain or in the sub-band frequency domain.
 
11. System according to one of the preceding claims, wherein the microphone array comprises at least one directional microphone.
 
12. Method for speech signal processing with combined noise reduction and echo compensation, comprising
receiving first speech signals as input signals for an adaptive beamforming employing adaptive beamforming filters,
receiving input signals for an adaptive compensating of acoustic echoes employing adaptive echo compensation filters,
determining combined adaptive filters from the adaptive beamforming filters and the adaptive echo compensation filters,
simultaneously adapting the adaptive beamforming filters and the adaptive echo compensation filters by adapting the combined adaptive filters, and
obtaining adaptively beamformed signals by the adaptive beamforming and echo compensating the adaptively beamformed signals to obtain an output signal of the speech signal processing.
 
13. Method according to Claim 12, further comprising
detecting microphone signals by a microphone array comprising at least two microphones,
obtaining loudspeaker signals comprising second speech signals output by at least one loudspeaker, and wherein
the first speech signals are obtained on the basis of the microphone signals, and
the input signals for the adaptive compensating of acoustic echoes are the loudspeaker signals.
 
14. Method according to Claim 12 or 13, wherein the adaptive beamforming is performed by an adaptive beamformer.
 
15. Method according to Claim 12 or 13, wherein the adaptive beamforming comprises
receiving and processing of the first speech signals by a blocking matrix to obtain noise reference signals,
processing of the noise reference signals by an adaptive noise canceling means to obtain adaptive noise signals, and
receiving and processing of the first speech signals by a beamformer to obtain first beamformed signals, and
subtracting the adaptive noise signals obtained by the adaptive noise canceling means from the first beamformed signals to obtain adaptively beamformed signals.
 
16. Method according to Claim 15, wherein adaptive filters are used for the adaptive echo compensation and the adaptive noise canceling and the filter length of the filter of the adaptive echo compensating means is equal to or longer than the filter length of the filter of the adaptive noise canceling means.
 
17. Method according to one of the Claims 12 ― 16, wherein the combined adaptive filters are adapted by applying a Least Square optimization criterion.
 
18. Method according to Claim 17, wherein the constraint for the Least Square optimization criterion reads


where the upper index T denotes the vector transposition, C is the conventional constraint matrix known from LCMV beamforming, w(k) and a(k) are the filters of the adaptive beamforming and the adaptive noise canceling, respectively, Q is the number of the at least one loudspeaker, Na is the number of filter coefficients of the filters of the adaptive echo canceling means and c is the constraint vector representing C constraints.
 
19. Method according to one of the Claims 13 - 18, the microphone signals are compensated for time delay to obtain the first speech signals.
 
20. Method according to one of the Claims 13 - 19, wherein the microphone signals or the time delay compensated microphone signals are adaptively self-calibrated to obtain the first speech signals.
 
21. The method of one of the Claims 12 - 20, wherein the adaptive beamforming and the adaptive echo compensating and/or the time delay compensating and/or the adaptive self-calibrating are performed in the time domain or in the frequency domain or in the sub-band frequency domain.
 
22. Computer program product, comprising one or more computer readable media having computer-executable instructions for performing the steps of the method according to one of the Claims 12 ― 21.
 




Drawing