WO2009034524A1

WO2009034524A1 - Apparatus and method for audio beam forming

Info

Publication number: WO2009034524A1
Application number: PCT/IB2008/053637
Authority: WO
Inventors: David A. C. M. Roovers; Valery S. Kot
Original assignee: Koninklijke Philips Electronics N.V.
Priority date: 2007-09-13
Filing date: 2008-09-09
Publication date: 2009-03-19

Abstract

An audio beam forming apparatus comprises a first sound sensor (201, 203, 205) having a first directional sensitivity and providing a first signal and a second sound sensor (201, 203, 207) having a second directional sensitivity and providing a second signal. The second directional sensitivity is in a different direction than the first directional sensitivity. A beam forming unit (100) combines at least the first and second directional signals to generate a combined directional signal. The beam forming unit (100) is arranged to adapt an audio beam direction for the combined directional signal in response to a desired sound source. The beam forming unit (100) may specifically be a Filter and Sum Beamformer and the directional input signals to the beam former may reduce and decorrelate correlated noise, such as reverberation noise, thereby improving beam forming performance. The directional input signals may specifically allow very close microphones to be used, such as in hearing aids.

Description

APPARATUS AND METHOD FOR AUDIO BEAM FORMING

FIELD OF THE INVENTION

The invention relates to audio beam forming and in particular, but not exclusively, to audio beam forming for a hearing aid apparatus.

BACKGROUND OF THE INVENTION

Advanced processing of audio signals has become increasingly important in many areas including e.g. telecommunication, content distribution etc. For example, in some applications, such as teleconferencing, complex processing of inputs from a plurality of microphones has been used to provide a configurable directional sensitivity for the microphone array comprising the microphones. Specifically, the processing of signals from a microphone array can generate an audio beam with a direction that can be changed simply by changing the characteristics of the combination of the individual microphone signals.

Typically, beam form algorithms are controlled such that the attenuation of interferers is maximized. For example, a beamforming algorithm can be controlled to provide a maximum attenuation (preferably a null) in the direction of a signal received from a main interferer.

A beam form algorithm which provides particularly advantageous performance in many embodiments, is the Filtered-Sum Beamformer (FSB) disclosed in Patent Cooperation Treaty patent application publication WO 99/27522. In contrast to many other beam formers, the FSB algorithm seeks to maximize the sensitivity of the microphone array towards a desired signal rather than to maximize attenuation towards an interferer. An example, of the FSB algorithm is illustrated in FIG. 1 The FSB algorithm seeks to identify the acoustic impulse responses from a desired source to an array of microphones, including the direct field and the first reflections. The FSB creates an enhanced output signal, z, by adding the desired part of the microphone signals coherently by filtering the received signals in forward matching filters and adding the filtered outputs. Also, the output signal is filtered in backward adaptive filters having conjugate filter responses to the forward filters (in the frequency domain corresponding to time inversed impulse responses in the time domain). Error signals are generated as the difference between the input signals and the outputs of the backward adaptive filters, and the coefficients of the filters are adapted to minimize the error signals thereby resulting in the audio beam being steered towards the dominant signal. The generated error signals can be considered as noise reference signals which are particularly suitable for performing additional noise reduction on the enhanced output signal z.

A particularly important area for audio signal processing is in the field of hearing aids. In recent years, hearing aids have increasingly applied complex audio processing algorithms to provide an improved user experience and assistance to the user. For example, audio processing algorithms have been used to provide an improved signal to noise ratio between a desired sound source and an interfering sound source resulting in a clearer and more perceptible signal being provided to the user. In particular, hearing aids have been developed which include more than one microphone with the audio signals of the microphones being dynamically combined to provide directivity for the microphone arrangement. As another example, noise canceling algorithms may be applied to reduce the interference caused by undesired sound sources and background noise.

The FSB algorithm promises to be advantageous for applications such as hearing aids as it promises an efficient beam forming towards a desired signal (rather than being directed to attenuation of interfering signals). This has been found to be of particular advantage in hearing aid applications where it has been found to provide a signal to the user which facilitates and aids the perception of the desired signal. In addition, the FSB algorithm provides a noise reference signal which is particularly suitable for noise reduction/ compensation for the generated signal.

However, it has been found that the FSB algorithm has some associated disadvantages when used in applications such as for a hearing aid. In particular, it has been found that for low distances between the microphones of the microphone array, the performance of the FSB algorithm degrades. For example, for a typically hearing aid configuration of an end- fire array with two omni-directional microphones with a spacing of 15 mm, the FSB has been found to have suboptimal performance. Indeed, it has been found that in many scenarios, the FSB algorithm has not been able to converge towards the desired signal.

Hence, an improved audio beam forming would be advantageous and in particular a beam forming allowing increased flexibility, facilitated implementation, improved suitability for hearing aids, increased accuracy, improved generation of noise reference signals, improved signal to noise ratio and/or improved performance would be advantageous.

SUMMARY OF THE INVENTION Accordingly, the Invention seeks to preferably mitigate, alleviate or eliminate one or more of the above mentioned disadvantages singly or in any combination.

According to an aspect of the invention there is provided audio beam forming apparatus comprising: a first sound sensor having a first directional sensitivity and providing a first signal; a second sound sensor having a second directional sensitivity and providing a second signal, the second directional sensitivity being in a different direction than the first directional sensitivity; a beam forming unit arranged to combine at least the first and second directional signals to generate a combined directional signal, the beam forming unit being arranged to adapt an audio beam direction for the combined directional signal in response to a desired sound source. The invention may provide improved beam forming in many scenarios and applications. In particular, the invention may be suitable for implementation in a hearing aid and/or may provide improved performance for hearing aid applications. Specifically, the invention may be suitable for implementation with a small distance between the microphones of the first and second sound sensors. The invention may in many scenarios provide improved beam forming in reverberant audio environments and/or with long distances between the apparatus and a desired sound source. The invention may allow improved beamforming thereby providing an improved combined signal e.g. having reduced signal to noise/interference ratio.

The inventors have in particular realized that a problem in many beam formers tracking a desired signal (such as the FSB algorithm) is that performance is degraded due to the noise signal at the microphones having increased correlation for smaller inter-microphone distances. The invention may reduce correlation between input signals to a beam form algorithm and may accordingly improve performance of the beamforming operation.

The invention may be particularly advantageous when used with a beam forming unit based on the FSB algorithm.

The beam forming unit may adapt the audio beam direction for the combined directional signal to have a main beam (with the maximum gain) directed in an angle of arrival of a dominant signal component for the desired sound source (which may specifically be a direct signal). The adaptation of the audio beam may be by adapting a weight or a filtering of the first and second signal prior to a summing of the resulting signals. Specifically, filter coefficients may be adapted to result in a coherent combination for signals from a desired direction.

The apparatus may in many embodiments be implemented with reduced complexity and/or may require reduced computational or other resource.

It will be appreciated that the audio beam forming apparatus may comprises additional sound sensors and that the beam forming unit may further combine these signals with the first and second directional signals to generate the combined directional signal. Specifically, the audio beam form apparatus may comprise two or more microphones and the beam form algorithm may be based on signals from all of the microphones.

The directional sensitivity may be such that at least one attenuation angle interval is at least 10 dB lower than the directional sensitivity of at least one non-attenuation angle interval.

In accordance with an optional feature of the invention, the beam forming apparatus further comprises: an audio activity detection unit for detecting audio activity of the desired sound source; and wherein the beam forming unit is arranged to modify the combination of the at least first and second directional signals to adapt the audio beam direction only when the audio activity detection unit indicates that the desired sound source is active. This may provide improved performance and may in particular allow an improved beamforming to be provided. The feature may result in an increased effect of using directional inputs to the beam forming unit as the detection ensures that the update will tend to be performed when a dominant desired signal is present and thus when the decorrelating effect of directional sound sensors is increased. The audio activity detection may specifically limit the update of the beam forming to only take place when a dominant signal is present from the desired source thus resulting in a dominant signal tracking beam forming algorithm, such as the FSB, tracking the desired signal.

In accordance with an optional feature of the invention, the first sound sensor comprises: a first substantially omni-directional microphone; a second substantially omnidirectional microphone ; and a first directional beamforming unit arranged to generate a first directional audio beam by combining signals from the first substantially omni-directional microphone and the second substantially omni-directional microphone into the first signal. This may provide facilitated implementation and/or improved beam forming in many scenarios. In particular, the feature may allow a reduced number of microphones and/or the use of simpler and cheaper omni-directional microphones. The feature may furthermore facilitate and/or allow dynamic variations of the level of directivity. For example, the first and second signals may be changed from being directional signals to being omni-directional signals depending on the user preferences and/or current scenario.

The substantially omni-directional sensitivity may have a variation of less than 5 dB for different directions.

In accordance with an optional feature of the invention, the second sound sensor comprises: the first substantially omni-directional microphone ; the second substantially omni-directional microphone ; and a second directional beamforming unit arranged to generate a second directional audio beam by combining signals from the first substantially omni-directional microphone and the second substantially omni-directional microphone into the second signal. This may provide facilitated implementation and/or improved beam forming in many scenarios. In particular, the feature may allow a reduced number of microphones and/or the use of simpler and cheaper omni-directional microphones.

The substantially omni-directional sensitivity may have a variation of less than 5 dB for different direction. In accordance with an optional feature of the invention, the second sound sensor has a substantially maximum sensitivity direction in a substantially opposite direction of a substantially maximum sensitivity direction of the first sound sensor .

The first and second sound sensors may have directional audio beams pointing in substantially opposite directions (say within 10 degrees). The sensitivity pattern for the two sound sensors may be approximately the same and the generated beams for the first and second sound sensors may specifically be symmetric around a central axis.

In accordance with an optional feature of the invention, the first and second sound sensors are arranged to generate a correlation between the first and second signal lower than a correlation from signals of two omni-directional sound sensors at the same location as the first and second sound sensor.

This may improve performance of the beam forming. In particular, the inventors have realized that performance of beam formers using closely located sound sensors may be improved by using directional techniques to reduce correlation. In accordance with an optional feature of the invention, a distance between a sound sensor element of the first sound sensor and a different sound sensor element of the second sound sensor is less than 5 centimeters.

The invention may provide improved performance for closely located sound sensor elements. The sound sensor elements may specifically be microphones. The sound sensor elements of the different sound sensors may be the same or overlapping, i.e. a sound sensor element may be used for both the first sound sensor and the second sound sensor. The distance between two different sound sensor elements where one is used by the first sound sensor and the other is used by the second sound sensor may thus be less than 5 cm. The invention has been found to be particularly advantageous for microphone spacings of 0.5 to 2 cm.

In accordance with an optional feature of the invention, the beam forming unit comprises: a first adaptive filter for filtering the first signal to generate a first filtered signal; a second adaptive filter for filtering the second signal to generate a second filtered signal; a summing unit for summing the first and second filtered signal to generate the combined directional signal; an adaptation unit for setting coefficients of the first adaptive filter and the second adaptive filter to generate the combined directional signal, the adaptation unit being arranged to maximize a power estimate for a desired sound source component of the combined directional signal. This may provide particularly advantageous beam forming performance and the effect of the directional sound sensors providing inputs to the beam forming may be particularly high since such a beamforming unit is particularly sensitive to correlation effects. The adaptation unit may maximize the power estimate while limiting a power amplification measure of the first and/or second adaptive filters. Specifically, the power estimate may be maximized under the constraint that the summed energy of the transfer functions of the filters has a substantially constant energy (e.g. that it corresponds to a desired value with a given accuracy of, say, ±2dB) at predefined frequencies.

In accordance with an optional feature of the invention, the beam forming unit further comprises: a third adaptive filter for filtering the combined directional signal to generate a third filtered signal, a transfer function of the third adaptive filter corresponding to a delay compensated substantially complex conjugated transfer function of the first adaptive filter ; a difference unit for determining a difference signal between the first signal and the third filtered signal; and wherein the adaptation unit is arranged to adapt the transfer function of the first adaptive filter to reduce the difference signal. This may provide particularly advantageous beam forming performance and the effect of the directional sound sensors providing inputs to the beam forming may be particularly high as such a beamforming unit is particularly sensitive to correlation effects. The adaptation unit may reduce the difference signal while limiting a power amplification measure of the first and/or second adaptive filters. Specifically, the difference signal may be reduced or minimized under the constraint that the impulse response of the filters has a constant energy. In the frequency domain, the filter coefficients of the third adaptive filter may correspond to complex conjugated filter coefficients of the first adaptive filter (possibly phase offset in response to a delay). The feature may further allow for the generation of a reliable noise reference signal with improved performance due to the directional sound sensor inputs to the beam forming unit.

In some embodiments, the beam forming unit further comprises: a fourth adaptive filter for filtering the combined directional signal to generate a fourth filtered signal, a transfer function of the fourth adaptive filter corresponding to a delay compensated substantially complex conjugated transfer function of the second adaptive filter; a difference unit for determining a second difference signal between the second signal and the fourth filtered signal; and wherein the adaptation unit is further arranged to adapt the transfer of the second adaptive filter to reduce the second difference signal. In accordance with an optional feature of the invention, the beam forming apparatus further comprises a noise reduction unit arranged to compensate the combined directional signal in response to the difference signal.

This may provide an improved output signal in the form of the compensated combined directional signal. Specifically, a signal to noise ratio of the output signal may be improved.

In accordance with an optional feature of the invention, the noise reduction unit comprises: an audio activity detection unit for detecting audio activity of the desired sound source; an adaptive noise filter for filtering the difference signal to generate a filtered difference signal; a subtraction unit for subtracting the filtered difference signal from the combined directional signal to generate a compensated signal; and a noise adaptation unit arranged to adapt an impulse response of the adaptive noise filter to reduce a power measure of the compensated signal only when the audio activity detection unit indicates that the desired sound source is not active. This may provide an efficient, practical and/or high performance noise reduction. In particular, a reduction of correlation between the input signals for the functionality generating the noise reference signal (the difference signal) may result in an improved noise reduction. In accordance with an optional feature of the invention, the beam forming apparatus further comprises controller arranged to adapt a level directionality of at least one of the first and second sensor in response to a measure of noise correlation between sound sensor elements of the first and second sound sensor .

This may allow improved beam forming and may in particular allow improved performance in dynamically changing audio environments. For example, the level of directionality may be increased when the correlation of the captured audio increases. The feature may allow a trade off between noise reduction for uncorrelated and correlated noise to be dynamically optimized.

The noise correlation may specifically be between sound sensor elements of the first and second sensor (such as the correlation of noise for signals from omni-directional microphone signals prior to a directional processing by the first and second sound sensor) and/or may be noise correlation between the first and second signal.

In accordance with an optional feature of the invention, the audio activity detection apparatus is comprised in a hearing aid apparatus. The invention may provide an improved hearing aid. In particular, an improved beam forming may provide improved audio processing in the hearing aid resulting in improved performance and clarity of the signal provided to the user. Also, the invention may in many embodiments allow improved noise reduction.

According to another aspect of the invention there is provided a method of audio beam forming comprising: a first sound sensor having a first directional sensitivity providing a first signal; a second sound sensor having a second directional sensitivity providing a second signal, the second directional sensitivity being in a different direction than the first directional sensitivity; a beam forming unit combining at least the first and second directional signals to generate a combined directional signal, the beam forming unit being arranged to adapt an audio beam direction for the combined directional signal in response to a desired sound source.

These and other aspects, features and advantages of the invention will be apparent from and elucidated with reference to the embodiment(s) described hereinafter. BRIEF DESCRIPTION OF THE DRAWINGS

Embodiments of the invention will be described, by way of example only, with reference to the drawings, in which

FIG. 1 is an illustration of a beam forming unit; FIG. 2 is an illustration of an example of a beam forming apparatus in accordance with some embodiments of the invention;

FIG. 3 illustrates an example of a cardioid audio beam form; FIG. 4 is an illustration of an example of a beam forming apparatus in accordance with some embodiments of the invention; FIG. 5 is an illustration of a beam forming architecture;

FIG. 6 illustrates an example of signal coherence for different beam forms; FIG. 7 illustrates an example of noise characteristics for a microphone array; and

FIG. 8 is an illustration of an example of a pre-processing stage for a beam forming apparatus in accordance with an embodiment of the invention.

DETAILED DESCRIPTION OF SOME EMBODIMENTS OF THE INVENTION

The following description focuses on embodiments of the invention applicable to a hearing aid and in particular to a hearing aid comprising two substantially omni- directional microphones. However, it will be appreciated that the invention is not limited to this application but may be applied to many other audio applications. In particular, it will be appreciated that the described principles may readily be extended to embodiments based on more than two (omni-directional) microphones.

In the example, the hearing aid comprises a beam forming unit which is arranged to adapt an audio beam towards a desired sound source which specifically may be a speaker with which the user of the hearing aid is currently talking. In the specific example, the hearing aid comprises a beamform unit 100 as shown in FIG. 1.

The beam form unit 100 of FIG. 1 receives two input signals ui,u₂ and processes these to generate an audio beamform. Conventionally, the input signals ui,u₂ are received from omni-directional microphones but in the exemplary embodiment the input signals ui,u₂to the beam forming unit 100 are provided by (at least) two directional sound sensors as will be described in more detail later.

In the beam forming unit 100, the first input signal ui is fed to a first adaptive filter 101 which generates a first filtered signal. The second input signal u₂ is fed to a second adaptive filter 103 which generates a second filtered signal. The first and second filtered signals are then summed in a summing unit 105 to generate a combined directional signal. The combined directional signal corresponds to a received audio signal from a sound sensor having a directional sensitivity. Specifically, by modifying the filter coefficients of the first and second adaptive filters 101, 103, the direction of an audio beam can be directed in a desired direction.

The filter coefficients are updated such that a power estimate for a desired sound source component of the combined directional signal is maximized. In the example, the filter coefficients are updated when a signal from a desired sound source is dominant and therefore the desired sound component is presumed to be the dominant source component of the combined directional signal. Accordingly, the filter coefficients are updated such that a power measure for the entire combined directional signal is maximized. A suitable power measure may for example be obtained by squaring (or taking the absolute value) of samples of the combined directional signal and filtering the result with a suitable low pass filter. The adaptation of the filter coefficients are furthermore made with a constraint that the summed energy of the transfer functions of the adaptive filters 101, 103 is maintained constant at predefined frequencies.

In the specific example, the filter coefficients 101, 103 are not adapted directly. Instead, the beam forming unit 100 furthermore comprises a third adaptive filter 107 for filtering the combined directional signal to generate a third filtered signal and a fourth adaptive filter 109 for filtering the combined directional signal to generate a fourth filtered signal.

The third filtered signal is fed to a first subtraction unit 111 which generates a first difference signal between the third filtered signal and the first input signal ui (delayed by a delay 113). The fourth filtered signal is fed to a second subtraction unit 115 which generates a second difference signal between the fourth filtered signal and the second input signal u₂ (delayed by a delay 117).

In the system, the filter coefficients of the adaptive filters 107, 109 are adapted in the presence of a dominant signal from the desired sound source such that the difference signals X₁, x₂ are reduced and specifically minimized. A suitable algorithm for doing so is the well known Normalized Least Mean Squares algorithm. Periodically, for example after each data block of N samples, the time reversed filter coefficients 107 are copied to the filter 101 and the time reversed coefficients 109 are copied to the filter 103. By doing so, the power of the output signal z in the presence of a dominant signal from the desired sound source is maximized by the beam forming unit 100.

In the frequency domain, the transfer function of the third adaptive filter 107 corresponds to a complex conjugate of a transfer function of the first adaptive filter 101 and the transfer function of the fourth adaptive filter 109 corresponds to a complex conjugate of a transfer function of the second adaptive filter 103.

In more detail, the operation of the beam forming unit 100 may be described with reference to a signal model which use frequency domain quantities that are the Fourier tranforms of continuous time signals. It is assumed that each of the microphone signals contains a signal component from a desired source component, a reverberant signal component, and noise.

According to the model, the vector of microphone signals u(ω) = [U₁ (ω) u₂ (ω)]' (where the superscript t denotes transposition) is given by

u(ω) = h(co)s(co) + d(ω) + n(ω)

where s(ω) is the desired source signal, h(ω) is the vector of acoustic impulse responses from the source to the microphones including the direct sound path and possibly some early reflections, d(ω) is reverberation, and n(ω) is uncorrelated noise having equal variance on each of the microphones. Assuming that the desired signal, reverberation and noise are mutually uncorrelated, the input cross-power spectral density is given by

P_MM(ω) = E{u(ω)u»} = σ>)h>)h'(ω) +σj(ω)P_ώ(ω) +σ ²(ω)I(ω)

where σ_s ² (ω) , σ^(ω) and σ^(ω) are the variances of the source signal, reverberation, and noise, respectively, l(ω) is the identity matrix, the superscript h denotes complex conjugate transposition and denotes complex conjugation.

The reverberation is modeled as a diffuse (spherically isotropic) sound field. Although this is a simplified theoretical model, it has been proven to be a valuable model in many applications and is useful to provide an insight into the operation and performance of the FSB algorithm used in the beam forming unit 100. For omni-directional microphones, the coherence matrix of the reverberation is given by

sin (ύd/c

P^(ω) = ωd/c sin (ύd/c „ ωd/c

where d is the distance between the microphones and c is the speed of sound.

Using the described signal model, the combined directional output signal z of the beam forming unit 100 is given by a filter and sum operation on the microphone signals:

z(ω) = f ' (ω)u(ω)

In the time domain the filter coefficients for the first and second adaptive filters 101, 103 are specifically time reversed copies of the filter coefficients of the third and fourth adaptive filters 107, 109 respectively. Thus, in the frequency domain the filter coefficients for the first and second adaptive filters 101, 103 are complex conjugated versions of the filter coefficients of the third and fourth adaptive filters 107, 109 respectively.

Generally, a delay will be introduced in order to ensure causality of the signal processing and accordingly the third and fourth adaptive filters 107, 109 are in the example phase compensated to take this delay into account. Also, the input signals U₁, u₂ are delayed in the delays 113, 117 prior to the generation of the difference signals. The delay τ of the delays 113, 117 are generally set equal to the length of the adaptive filters 101, 103, 107, 109.

It can be shown that the adaptive beam forming unit 100 of FIG. 1 maximizes the power of the combined directional output signal z under the constraint that the summed power of the filter transfer functions equals unity for all frequencies, which can be expressed mathematically as | f (ω) |²= 1 Vω . Thus, by setting the filter coefficients to minimize the difference signals X₁, x₂, the power of the combined directional output signal z is maximized thereby providing an adaptation of the audio beam towards the dominant audio signal.

Using the signal model and applying the constraint, the combined directional output signal z is given by:

^(ω) = E{f»P_MM(ω)f(ω)}

= σ ²(ω) I f (ω)h(ω) |² +σ ²(ω)f>)P^(ω)f(ω) + σ_B ²(ω) For the desired source alone (i.e. in the absence of reverberation and noise), the optimal coefficients are given by:

h» f _r (CO) = α (CO) - h(co) |

where α(co) is an arbitrary all-pass term.

This expression shows that the optimal filter coefficients are equal to the conjugated transfer functions of the desired source to each of the microphones, (disregarding a common unknown amplitude and phase factor). Due to the limited filter length of practical filters, the beam forming unit will in practice only estimate the first part of the time domain impulse responses (typically the part including the direct field and possibly some early reflections).

The inventors of the current invention have realized that a conventional application of the beam forming unit of FIG. 1 is disadvantageous for small microphone spacings. Furthermore, the inventors have realized that the degradation of the FSB algorithm applied is due to the increased correlation between microphones with reduced distance and that this increased correlation not only reduces performance but introduces a bias and may prevent the FSB algorithm from converging towards a desired sound source as will be described in the following.

Firstly, the presence of uncorrelated noise will be considered. It is noted that the power of the combined directional output signal z due to the noise n(ω) does not depend on the filter coefficients as long as the constraint on their norm is satisfied. It follows that the optimal coefficients are independent of the noise and specifically the variance σ^(co) . This implies that the beam forming unit 100 using an FSB algorithm has an inherent robustness in the presence of uncorrelated, equal variance noise on the microphones. Thus, the presence of uncorrelated noise is unlikely to result in the beam forming algorithm not converging towards the desired beam form/ direction.

However, the inventors have realized that the described algorithm is significantly degraded in the presence of correlated noise and therefore is unsuitable for many applications where close microphones are used resulting in increased correlated noise, such as reverberation noise. Specifically, as will be shown in the following, the inventors have realized that the presence of correlated noise may result in the algorithm converging towards suboptimal filter coefficients corresponding to suboptimal beam forms/directions or may result in the algorithm not converging.

Initially it is noted that if the microphone distance is sufficiently large the microphone correlation due to the diffuse field is low in the frequency range of interest thus resulting in the reverberation component d(ω) of the signal model mainly relating to uncorrelated noise (specifically the sine function in the previously provided coherence matrix for the reverberation quickly reduces to values close to zero). However, for a small microphone spacing, d, it can be shown that there is significant correlation between the microphones over a wide frequency range resulting in the reverberation component d(ω) of the signal model representing a significant level of correlated noise.

Considering an input signal to the beam forming unit consisting in only the reverberation component ά(ω), power of the combined directional signal z is maximized if the coefficient vector f(ω) equals the normalized eigenvector corresponding to the largest eigenvalue of P^(ω) . The normalized eigenvectors are real- valued and can be determined

from the coherence matrix of the reverberation as e_{ = [— \2, — V2]' and e₂ = [— V2, — V2]'

, respectively.

For the frequencies of interest, where correlation is significant, the first eigenvector e_\ has the largest eigenvalue thereby suggesting that simple summation of the input signals U₁, u₂ maximizes the power of the combined directional signal. Thus, if only the reverberation component is considered, the impulse responses of the first and second adaptive filter 101, 103 reduce to a simple impulse. This is clearly different from the optimal coefficients determined for the input signal originating from the desired sound signal and clearly is not the optimal solution for a microphone array in end-fire position relative to the desired source.

Thus, as realized by the inventors, for an input signal comprising a desired signal component, an uncorrelated noise component and a correlated noise component, the uncorrelated noise component will merely increase the variance of the generated filter coefficient estimates but will not introduce a bias to the estimates whereas the correlated noise will tend to bias the adaptation away from the correct values of the filter coefficients. Specifically, it has been found that for a small microphone array in a reverberant room, the reverberation may completely prevent the beam forming unit 100 from converging towards the correct solution. This is especially the case if the level of the reverberation is equal to, or larger than, the direct sound including early reflections, i.e. if the distance between the source and the microphones exceeds the reverberation radius. Of course, such a situation is typically the case for hearing aid applications wherein the distance between the microphones is low whereas the distance to the desired sound source (e.g. a speaker) is much larger. The inventors have further realized that the performance may be substantially improved by using input signals to the beam forming unit which are not received from omnidirectional microphones but are received from directional sound sensors with different directional sensitivities. Thus, in the described embodiments, the input signals to the (FSB) beam forming unit 100 are not omni-directional signals but are directional signals. For example, the first and second microphone 119, 121 may be directional microphones angled in different directions or may be omni-directional microphones mounted in an arrangement introducing a directional characteristic (for example a directional audio shielding of the microphones may be used).

The use of directional sound sensors angled in different directions results in a reduction in the correlation of the received noise, and specifically the microphone correlation due to a diffuse sound field can be substantially decreased thereby providing improved performance of the beam former. Thus, the directional input signals (e.g. provided by directional microphones) for the beam forming unit 100 has a correlation which is lower than the correlation which would be obtained for signals from two omni-directional sound sensor elements (e.g. omni-directional microphones) at the same location as the sound sensors (e.g. at the location of the directional microphones).

It should be noted that in contrast to a beam forming algorithm merely seeking to reduce the interference level e.g. by steering a null towards a main interferer, the use of directional sound inputs is particularly advantageous for a beam forming algorithm based on tracking a desired source. Specifically, for an interference canceling beam forming algorithm the update is typically performed when the desired signal is not active and therefore the reverberation is likely to be less significant and uncorrelated noise will tend to dominate over correlated noise. Furthermore, correlated noise is less likely to degrade the beam forming algorithm which seeks to reduce a power estimate (for the interferer) rather than to maximize this. In contrast, the desired signal tracking beam forming algorithm typically adapts in a different audio environment where the desired sound source is dominant and the presence of reverberation and correlated noise is highly significant. Furthermore, the desired signal tracking algorithm seeks to maximize the power measure for the desired signal rather than to reduce a power estimate for an interferer. Furthermore, as the inventors have realized and as explained above, a beam forming unit such as the one of FIG. 1 is particularly sensitive to correlated noise as this tends to introduce a bias rather than merely degrade the variance of the filter coefficient estimates. Specifically, typical beam forming algorithms are optimized to minimize the variance of the estimates and therefore the total noise level of the input signals is minimized when designing these. Thus, whereas the introduction of directional input signals is disadvantageous or counter intuitive in typical (interference canceling) beam formers because the uncorrelated noise level and the total noise level may be increased thereby, it is highly advantageous for the beam forming unit 100 of FIG. 1 as the reduction of correlated noise at the expense of increased levels of uncorrelated noise reduces the bias of the filter coefficient estimates (but possibly at the cost of an increased variance). Specifically, the introduction of directional input signals for the beam forming unit of FIG. 1 may allow improved convergence performance for very closely spaced input sound sensor elements.

In the following, an embodiment will be described which uses a beam form pre-processing to generate two directional sound source signals based on microphone inputs from substantially omni-directional microphones. FIG. 2 illustrates a beam forming apparatus in accordance with the embodiment.

In the example, the apparatus is implemented in a hearing aid to be worn on the ear of a user. The hearing aid comprises a first and second microphone 201, 203 which in the example are omni-directional microphones mounted in an end-fire configuration

(mounted along a line towards the front when worn by a user). As the microphones 201, 203 are mounted in a hearing aid to be worn on an ear, the distance between the microphones 201, 203 must inherently be kept very low resulting in the noise correlation between the microphones 201, 203 being relatively high and especially in the presence of significant reverberation originating from a desired sound signal.

It will be appreciated that an omni-directional microphone is a microphone for which the sensitivity variation as a function of the angle between a sound source and a reference direction is less than a given value. A microphone may for example be considered an omni-directional microphone if the difference between the highest and lowest signal level recorded for the same sound source at different angles is less than a threshold, which e.g. may be 1, 5 or 10 dB.

The first and second microphones 101, 103 are coupled to a first pre-beam form unit 205. The first pre-beam form unit 205 is arranged combine the signals recorded by the microphones 201, 203 to generate an output signal corresponding to a beamformed sensitivity pattern as will be known to the person skilled in the art.

In the specific example, the first pre-beam form unit 205 is arranged to combine the audio signals from the first and second microphone 101, 103 to generate a directional sensitivity corresponding to a forward facing cardiod beam pattern. An example of a cardioid beam pattern is shown in FIG. 3.

In more detail, in the example, a hearing aid with two omni-directional microphones spaced d meters apart in an end- fire configuration (oriented toward 0 degrees) is considered. The origin O is defined to be the center of the end- fire array. It is assumed that the point sources under consideration are at ear- level and thus at an elevation of around zero degrees. Considering a point source s(n,θ) sampled at time instant n and located at θ deg., where θ is the angle between the y-axis and a ray from the origin to the source, the output of the forward cardioid can be given by:

(n) ~ -(l + cosQ)s(n,Q) ≡ c_θs(n,θ),

where c_θ = — (1 + cosθ) is the response of a cardioid facing 0 degrees to a signal incident on

the array at θ deg., normalized to have unit response to a signal arriving from 0 degrees. The first and second microphones 101, 103 are furthermore coupled to a second pre-beam form unit 207. The second pre-beam form unit 207 is also arranged combine the signals recorded by the microphones 201, 203 to generate an output signal corresponding to a beamformed sensitivity pattern as will be known to the person skilled in the art.

In the specific example, the second pre-beam form unit 207 is arranged to combine the audio signals from the first and second microphone 101, 103 to generate a directional sensitivity corresponding to a backwards facing cardiod beam pattern. Thus, the first pre-beam form unit 205 and the second pre-beam form unit 207 generates directional output signals which have different associated directions and which in the specific example are in opposite directions. Thus, a substantially maximum sensitivity direction for the beam generated by the first pre-beam form unit 205 is in the example in a substantially opposite direction of a substantially maximum sensitivity direction of the second pre-beam form unit 207. An example of a suitable beamforming algorithm is for example described in G. W. Elko, "Superdirectional microphone arrays", ch. 10, pp. 181-238, in Acoustic Signal Processing for Telecommunications, S. L. Gay and J. Benesty, Eds. Kluwer Academic Publishers, 2000. It will be appreciated that although the pre-beam forming is described with reference to a first pre-beam form unit 205 and second pre-beam form unit 207, the functionality may be implemented in a single unit and accordingly the pair of the first pre- beam form unit 205 and second pre-beam form unit 207 will henceforth be referred to as the pre-beam form unit 205, 207. It will also be appreciated that in other embodiments, other beam patterns than a cardiod beam pattern may be used and that possibly the two beam patterns of the first pre- beam form unit 205 and second pre-beam form unit 207 may be different in not only direction but also in shape.

Thus, in the example, the first pre-beam form unit 205 and second pre-beam form unit 207 generates two directional signals which are fed to a beam form processor which tracks a desired signal. In the specific example, the beam form processor is the beam forming unit 100 of FIG. 1. Thus, the FSB beam form algorithm of the beam forming unit 100 is provided with directional input signals resulting in improved performance as described previously. Specifically, the first pre-beam form unit 205 and second pre-beam form unit 207 generates directional signals which has a correlation that is lower than the correlation of the original signals of the two omni-directional microphones 201, 203. Accordingly, the biasing of the beam form algorithm may be reduced leading to improved convergence performance.

The use of pre-beam forming based on substantially omni-directional microphones provides several advantages in many embodiments. For example, the pre-beam forming can be modified and therefore the characteristics of the directional signals can easily be modified in order to optimize performance. For example, the beam pattern(s) and or degree of directionality can easily be modified. Also, the system allows the apparatus to use simple, low-cost omni-directional microphones 201, 203 thereby reducing complexity and cost of the hearing aid. Furthermore, the use of omni-directional microphones allows the hearing aid to easily switch to conventional operation (e.g. using omni-directional signals for the beam forming or switching the beam forming off completely).

In some embodiments, the beam form apparatus comprises further functionality as will be described in the following with reference to FIG. 4 which shows additional features of the beam form apparatus of FIG. 2. In the apparatus of FIG. 4 the update of the beam forming unit is only performed when the received signal from the desired sound source is the dominant signal component in the received signal.

Specifically, the beam form apparatus comprises an audio activity detector 401 which is arranged to detect audio activity for the desired sound source. The audio activity detector 401 is coupled to the beam forming unit 100 and provides a control signal indicating whether audio activity of the desired signal is detected or not. Furthermore, the beam forming unit 100 is arranged to only adapt the beam form algorithm (e.g. the filter coefficients in the case of the beam forming unit of FIG. 1) when the control signal indicates that the desired sound source is active. Thus, the audio beam direction is only adapted/changed when the apparatus estimates that the desired signal is active and in the specific example only when the desired signal is the dominant signal.

By only updating the beam form characteristics when the desired signal is dominant an improved tracking of the desired sound signal is achieved resulting in improved performance. However, the correlation of noise tends to be high when the desired signal is dominant due to high levels of reverberation and as has been demonstrated previously such noise may result in a bias in the adaptation. Accordingly, the decorrelation of the input signals due to the use of directional sound sources interacts efficiently with the restriction of the adaptation to situations with a dominant desired signal to provide a substantially improved performance as it reduces correlation exactly when this is most important. Thus, the combination of the targeting of the adaptation to a desired signal and the use of directional filtering allows the FSB algorithm to be used with closely spaced microphones. Accordingly, even for small spacings of 05. - 1.5 cm as is typically used in hearing aids, a convergence of the FSB algorithm can be achieved despite the adaptation being limited to situations where the noise is much more likely to be closely correlated.

It will be appreciated that any suitable algorithm for detecting the presence of a desired signal may be used. In the specific example, the audio activity detector 401 is arranged to detect the dominance of a desired sound source by comparing the signal level in the forward facing directional audio beam (i.e. the output signal of the first pre-beam form unit 205) to the signal level from one of the omni- directional microphones. If the signal level in the forward facing beam is much higher than for an omni-directional beam, it is likely that a dominant signal is received from a forward direction. Furthermore, the normal use of a hearing aid, e.g. during conversation, means that it is likely that such a signal is a desired signal (e.g. from a person with which the user is speaking). In contrast, if the signal levels are similar or the omni-directional signal level is higher, no dominant sound source exists in the forward direction meaning that it is significantly less likely that the received signal has a dominant desired signal.

It will be appreciated that other, and specifically more complex, audio detection algorithms may be used without subtracting from the invention.

The beam form apparatus of FIG. 4 furthermore comprises a noise reduction functionality which can reduce the noise on the combined directional signal z being generated by the beam forming unit 100.

Specifically, as previously described, the adaptive filters 107, 109 of the beam forming unit 100 are adapted in the presence of a dominant desired signal such that a power measure of each of the difference signals X₁, x₂ is minimized. Thus, when the beam forming unit 100 has fully converged towards the desired sound source, the difference signals X₁, x₂ ideally do not contain any desired sound. However, they do contain undesired (noise) signals and therefore they are commonly called noise reference signals. Accordingly, these noise reference signals can be used to reduce the noise in the combined directional signal z. Accordingly, the noise reduction functionality reduces the noise of the combined directional signal by compensating the combined directional signal for one of the difference signals.

As explained previously, the use of directional input signals to the beam forming unit 100 of FIG. 1 not only provides an improved beamformed output signal but also substantially improves the noise reference signals in the form of the difference signals X₁, x₂ . Indeed, in practice it has been found that the improved quality of the noise reference signals allows a substantially improved noise reduction to be performed resulting in a substantially improved output signal. In the specific example, the first difference signal X₁ is fed to an adaptive noise filter 403 which in the specific example is a time domain Finite Impulse Response (FIR) filter. The adaptive noise filter 403 generates a filtered difference signal which is fed to a subtraction unit that subtracts the filtered difference signal from the combined directional signal from the beam forming unit 100. The output of the subtraction unit 405 is fed to a noise adaptation unit 409 which is coupled to the adaptive noise filter 403. The noise adaptation unit 409 is arranged to adapt the impulse response of the adaptive noise filter 403.

When the desired sound source is not active, the output of the subtraction unit 405 is ideally zero corresponding to all noise being cancelled. Thus, any signal component remaining in this scenario is due to noise or interference picked up by the microphones 201, 203 and not being completely cancelled. Accordingly, the noise adaptation unit 409 seeks to adapt the filter coefficients to minimize a power measure for the output signal y when the desired signal is not active. Accordingly the noise adaptation unit 409 is also coupled to the audio activity detector 401 and is arranged to only adapt the adaptive noise filter 403 when the indication from the audio activity detector 401 indicates that the desired signal is not active.

It will be appreciated that the criterion for adapting the adaptive noise filter 403 is not necessarily the complementary of the criterion for updating the beam forming. For example, the beam forming unit 100 may only be updated if it is estimated that the desired signal is dominant whereas the adaptive noise filter 403 may only be adapted if it is estimated that there is no desired signal component present at all. In the specific example, different thresholds for the ratio between the signal level in the forward beam and the signal level in the omni-directional beam may be applied. For example, the beam forming may only be adapted if the ratio exceeds, say, 10 dB whereas the adaptive noise filter 403 is only adapted if the ratio is less than, say 0 dB.

The noise adaptation unit 409 may for example adapt the filter coefficients of the adaptive noise filter 403 using the normalized least mean square algorithm described in S. Haykin, "Adaptive Filter Theory", Ch. 6, 3^rd ed. Prentice Hall, 1995. The use of directional input sensors to the beam forming unit 100 thus not only allows the beam forming algorithm to converge and generate an improved desired signal but also allows improved noise reference signals to be generated which can be used to effectively reduce the noise components of the output signal.

Specifically, the combination of the preprocessing, the desired signal tracking beam forming algorithm and a noise cancelling stage using the difference signals/ noise reference signals results in a beamforming with significantly improved (and typically optimal) desired signal quality and high noise reduction.

It should be noted that conventional interference cancelling beam formers make an assumption of the direction of the desired sound source and then use a simple geometrical model for estimating the impulse responses from the desired source to the microphones (i.e. a model based on pure acoustic travel time differences depending on the distance from the source to each of the microphones is typically applied). The inaccuracy of such a model typically results in distortion of the desired speech by the beam former. Thus, an interference tracking beam former reduces interference but also distorts or cancels part of the desired signal. The desired signal tracking beam former, and in particular the FSB algorithm provides much more reliable estimates of the impulse responses from the desired sound source to the microphones, even in the presence of reflections, and the estimates may be used to provide improved noise reference signals (i.e. not containing any desired signal). As a result, the noise reduction process following the beam forming unit 100 does not distort the desired signal.

In the following the reduction of correlation achieved by applying directional microphones will be demonstrated.

Firstly it will be appreciated that directional microphones may be created from omni-directional microphones in various ways including using e.g. a differential array or an optimal beam former implemented e.g. as an Linearly Constrained Minimum Variance (LCMV) beamformer or as a Generalized Sidelobe Canceller.

In the following, a pre-beam forming approach as illustrated in FIG. 5 will be considered. In this beam former forward and backward facing responses are created using filter vectors v_f (ω) and v_B(co) , respectively, producing corresponding output signals x_P(θύ) and x_B (ω) :

X_p (co) = v^, (co)u(co) and x_B (co) = Ύ_B' (CO)U(CO)

Considering the example of differential microphones, the filter vectors can be given by:

v_F (ω) = [ 1 - e^jayτ Y and Y_B (co) = [-e^jayτ 1 ]'

In these expressions, the delay τ determines the direction of minimum sensitivity of a differential microphone relative to its forward direction.

The coherence between the output signals x_F(ϋi) and X₈ (co) of the forward and backward facing audio beams can be determined as:

E{x_f (co)^*x_B (co)}

C(ω) =

VE{| x_F(ω) |²}E{| x_B(ω) |²}

For the diffuse field, this can be rewritten as ₌ v£ (CO)P₁₆, (ω)v» v^ (^Q)P (^Q) v* (ω)

For differential processing it can be shown that:

The squared magnitude of this function is plotted in FIG. 6 for various values

■^■£ of α = . It can be seen that all configurations significantly reduce the coherence, τ + d/c while the so-called super-cardioid configuration results in zero coherence across the entire frequency range.

Thus, it is shown that the directional preprocessing transforms the reverberation into an signal with reduced correlation and potentially reduced variance. Thus, as uncorrelated noise does not bias the adaptation away from the correct values, a much improved performance is achieved and in particular the directional input signals can substantially improve FSB convergence in the presence of reverberation.

Introducing directional pre-processing has resulted in the reverberation signal ά(ω) being both reduced and decorrelated. However, as a consequence of the pre-processing the uncorrelated noise n(ω) may be increased and some correlation may also be introduced to previously uncorrelated noise.

FIG. 7 illustrates noise characteristics for a supercardioid beamform. As illustrated by the example, the diffuse (correlated) noise is reduced by several dBs whereas the uncorrelated noise is increased by up to 15 dB. Moreover, FIG. 7 shows that uncorrelated noise at the microphone inputs can be transformed into signals with coherence. Hence, the beam form convergence may be more sensitive to uncorrelated noise such as microphone self noise and electronic noise. However, in reverberant environments, the level of reverberation and diffuse noise is usually much higher than the uncorrelated noise level. Moreover, in contrast to the reverberation, the uncorrelated microphone noise is always stationary, making it easy to detect it and to stop the adaptation of the beam form algorithm accordingly. In some embodiments, the beam forming apparatus may as illustrated in FIG. 8 furthermore comprise a controller unit 801 which determines a measure of the correlation of the noise from the microphones 201, 203. The controller unit 801 may e.g. continuously determine a cross correlation for the microphone signals U₁, u₂ during times when the audio activity detector 401 detects that no desired signal is present. If the correlation exceeds a given threshold, the controller unit 801 may control the beam forming unit 100 to stop adapting the filter coefficients.

In some embodiments, a controller unit 801 is furthermore coupled to the pre- beam form unit 205, 207 (corresponding to the first and second pre-beam form units 205, 207) and is arranged to control a level of the directionality for the audio beams generated by the beam form unit 205, 207. Specifically, as the correlation increases the beams are made narrower and as the correlation decreases the beams are made broader. In a simple embodiment, the pre-beam form unit 205, 207 may simply switch between directional and omni-directional beams depending on whether the noise correlation exceeds a threshold or not.

It will be appreciated that the above description for clarity has described embodiments of the invention with reference to different functional units and processors. However, it will be apparent that any suitable distribution of functionality between different functional units or processors may be used without detracting from the invention. For example, functionality illustrated to be performed by separate processors or controllers may be performed by the same processor or controllers. Hence, references to specific functional units are only to be seen as references to suitable means for providing the described functionality rather than indicative of a strict logical or physical structure or organization. The invention can be implemented in any suitable form including hardware, software, firmware or any combination of these. The invention may optionally be implemented at least partly as computer software running on one or more data processors and/or digital signal processors. The elements and components of an embodiment of the invention may be physically, functionally and logically implemented in any suitable way. Indeed the functionality may be implemented in a single unit, in a plurality of units or as part of other functional units. As such, the invention may be implemented in a single unit or may be physically and functionally distributed between different units and processors.

Although the present invention has been described in connection with some embodiments, it is not intended to be limited to the specific form set forth herein. Rather, the scope of the present invention is limited only by the accompanying claims. Additionally, although a feature may appear to be described in connection with particular embodiments, one skilled in the art would recognize that various features of the described embodiments may be combined in accordance with the invention. In the claims, the term comprising does not exclude the presence of other elements or steps. Furthermore, although individually listed, a plurality of means, elements or method steps may be implemented by e.g. a single unit or processor. Additionally, although individual features may be included in different claims, these may possibly be advantageously combined, and the inclusion in different claims does not imply that a combination of features is not feasible and/or advantageous. Also the inclusion of a feature in one category of claims does not imply a limitation to this category but rather indicates that the feature is equally applicable to other claim categories as appropriate. Furthermore, the order of features in the claims do not imply any specific order in which the features must be worked and in particular the order of individual steps in a method claim does not imply that the steps must be performed in this order. Rather, the steps may be performed in any suitable order. In addition, singular references do not exclude a plurality. Thus references to "a", "an", "first", "second" etc do not preclude a plurality. Reference signs in the claims are provided merely as a clarifying example shall not be construed as limiting the scope of the claims in any way.

Claims

CLAIMS:

1. An audio beam forming apparatus comprising: a first sound sensor (201, 203, 205) having a first directional sensitivity and providing a first signal; a second sound sensor (201, 203, 207) having a second directional sensitivity and providing a second signal, the second directional sensitivity being in a different direction than the first directional sensitivity; a beam forming unit (100) arranged to combine at least the first and second directional signals to generate a combined directional signal, the beam forming unit being arranged to adapt an audio beam direction for the combined directional signal in response to a desired sound source.

2. The beam forming apparatus of claim 1 further comprising: an audio activity detection unit (401) for detecting audio activity of the desired sound source; and wherein the beam forming unit (100) is arranged to modify the combination of the at least first and second directional signals to adapt the audio beam direction only when the audio activity detection unit indicates that the desired sound source is active.

3. The beam forming apparatus of claim 1 wherein the first sound sensor (201, 203, 205) comprises: a first substantially omni-directional microphone (201); a second substantially omni-directional microphone (203); and a first directional beamforming unit (205) arranged to generate a first directional audio beam by combining signals from the first substantially omni-directional microphone (201) and the second substantially omni-directional microphone (203) into the first signal.

4. The beam forming apparatus of claim 3 wherein the second sound sensor (201, 203, 207) comprises: the first substantially omni-directional microphone (201); the second substantially omni-directional microphone (203); and a second directional beamforming unit (207) arranged to generate a second directional audio beam by combining signals from the first substantially omni-directional microphone (201) and the second substantially omni-directional microphone (203) into the second signal.

5. The beam forming apparatus of claim 1 wherein the second sound sensor (201, 203, 207) has a substantially maximum sensitivity direction in a substantially opposite direction of a substantially maximum sensitivity direction of the first sound sensor (201, 203, 205).

6. The beam forming apparatus of claim 1 wherein the first and second sound sensors (201, 203, 205, 207) are arranged to generate a correlation between the first and second signal lower than a correlation from signals of two omni-directional sound sensors at the same location as the first and second sound sensor.

7. The beam forming apparatus of claim 1 wherein a distance between a sound sensor element of the first sound sensor (201, 203, 205) and a different sound sensor element (201, 203, 207) of the second sound sensor is less than 5 centimeters.

8. The beam forming apparatus of claim 1 wherein the beam forming unit (100) comprises: a first adaptive filter (101) for filtering the first signal to generate a first filtered signal; a second adaptive filter (103) for filtering the second signal to generate a second filtered signal; a summing unit (105) for summing the first and second filtered signal to generate the combined directional signal; an adaptation unit for setting coefficients of the first adaptive filter and the second adaptive filter to generate the combined directional signal, the adaptation unit being arranged to maximize a power estimate for a desired sound source component of the combined directional signal.

9. The beam forming apparatus of claim 8 wherein the beam forming unit (100) further comprises: a third adaptive filter (107) for filtering the combined directional signal to generate a third filtered signal, a transfer function of the third adaptive filter (107) corresponding to a delay compensated substantially complex conjugated transfer function of the first adaptive filter (101); a difference unit (111) for determining a difference signal between the first signal and the third filtered signal; and wherein the adaptation unit is arranged to adapt the transfer function of the first adaptive filter (101) to reduce the difference signal.

10. The beam forming apparatus of claim 9 further comprising a noise reduction unit (401, 403, 405, 409) arranged to compensate the combined directional signal in response to the difference signal.

11. The beam forming apparatus of claim 10 wherein the noise reduction unit (401, 403, 405, 409) comprises: an audio activity detection unit (401) for detecting audio activity of the desired sound source; an adaptive noise filter (403) for filtering the difference signal to generate a filtered difference signal; a subtraction unit (405) for subtracting the filtered difference signal from the combined directional signal to generate a compensated signal; and a noise adaptation unit (409) arranged to adapt an impulse response of the adaptive noise filter (403) to reduce a power measure of the compensated signal only when the audio activity detection unit (401) indicates that the desired sound source is not active.

12. The beam forming apparatus of claim 1 further comprising a controller (801) arranged to adapt a level directionality of at least one of the first and second sensor (201, 203,

205, 207) in response to a measure of noise correlation between sound sensor elements of the first and second sound sensor (201, 203, 205, 207).

13. A hearing aid comprising the beam forming apparatus of claim 1.

14. A method of audio beam forming comprising: a first sound sensor (201, 203, 205) having a first directional sensitivity providing a first signal; a second sound sensor (201, 203, 207) having a second directional sensitivity providing a second signal, the second directional sensitivity being in a different direction than the first directional sensitivity; a beam forming unit (100) combining at least the first and second directional signals to generate a combined directional signal, the beam forming unit being arranged to adapt an audio beam direction for the combined directional signal in response to a desired sound source.