EP3065417B1 - Method for suppressing interference noise in an acoustic system - Google Patents

Description

The invention relates to a method for suppressing a noise in an acoustic system, wherein the acoustic system comprises at least one microphone and at least one speaker, and wherein the at least one microphone generates an input signal, wherein the at least one loudspeaker generates an acoustic signal, which partially the at least one microphone feeds back.

In an acoustic system of the type described above, such as may be provided by a hearing aid, noise may occur due to feedback. Acoustic feedback may result from the fact that the acoustic signal generated by the speaker is partially perceived by the microphone, and thereby re-entry into the acoustic system. The input signal generated by the microphone is amplified in the acoustic system, so that within the closed loop, which is formed by the speaker, the acoustic signal generated by this, the microphone, and the signal processing within the acoustic system, a signal component by the feedback continues is amplified to a whistling noise when the gain in the signal processing within the acoustic system exceeds a certain limit.

Such noise can be reduced or even eliminated by so-called feedback cancellers. For this purpose, the prior art often uses adaptive feedback cancellation methods in which an adaptive filter with filter coefficients h models the time-dependent impulse response of the acoustic feedback path. An often used example of a rule for adjusting the filter coefficients h is given by the "normalized least mean square" algorithm (NLMS): $$\mathrm{H}\left(\mathrm{k}+1\right)=\mathrm{H}\left(\mathrm{k}\right)+\mathrm{\mu \; e}*\left(\mathrm{k}\right)\mathrm{x}\left(\mathrm{k}\right)/{\left|\mathrm{x}\left(\mathrm{k}\right)\right|}^2,$$

Where k is the discrete time index, x is the input to the feedback cancellation system, e = m-c is the error signal, which is defined as the difference between the microphone-generated input m and the compensation signal c to compensate for the feedback. μ is the step size over which the speed of adaptation or convergence is controlled, and * denotes the complex conjugation.

In a realistic acoustic system, the input signal m is often first digitized with a comparatively high sampling rate and thereby converted into time-discrete sampled values. Subsequently, in each case a multiplicity, for example 128, of consecutive samples is combined to form a so-called frame. Within a frame, a spectral analysis of the input signal can now be carried out on the basis of the samples forming the frame by means of Fourier transformation. For the generation or the analysis of a next frame, the window to be considered is shifted in the direction of the time axis by a few sample values, for example 32, so that the windows of the respective samples to be taken into account for one frame overlap partially for adjacent frames. The time index can in this case be understood as a frame index, whereby the adaptive filter can also be used in the frequency domain. In this case, the filter coefficients h are vectors whose entries each correspond to a spectral subband. However, the application is not limited to this case. Details can be found for example in S. Haykin, "Adaptive Filter Theory" (Englewood Cliffs, NJ: Prentice-Hall, 1996 ) or T. v. Waterschoot & M. Moonen, "Fifty years of acoustic feedback control: state of the art and future challenges "(Proc. IEEE, Vol. 99, No. 2, Feb. 2011, pages 288-327 ).

It is now a known problem that correlated input signals, such as may be generated by the recording of music or spoken language, may result in divergence in an adaptive filter, which may result in at least partial cancellation of a target signal. This can produce noticeable signal artifacts in the output signal, resulting in a significant deterioration of the sound quality. The whistling noise generated by acoustic feedback also has a high correlation in the signals concerned, particularly when there is a correlated target signal which is picked up and fed back through a loudspeaker after playback. If an adaptive filter is to be used to suppress the interference noise generated thereby, signal components of the target signal can also be at least partially extinguished in the suppression of the interference signal of the feedback, which has a negative effect on the sound quality of the output signal.

From the EP 2 503 795 A2 a method for operating a binaural hearing device with a feedback suppression device is known, in which a pre-whitening filter of a first hearing device is controlled in response to a signal of a second hearing device.

In the DE 2013 207 403 B3 a method for controlling an adaptation step size of an adaptive filter of a hearing device is described in which an output signal for driving an adaptive filter is frequency-shifted. In this case, an autocorrelation value of an input signal is used as a function of the frequency shift for controlling the adaptation step size.

The EP 2 736 271 A1 discloses a system for estimating a feedback noise signal in which an output signal is shifted in frequency. The frequency-shifted output signal is then supplied as an input to an adaptive filter for rejecting a noise signal and to an algorithm for adjusting the filter parameters of the adaptive filter.

The invention is therefore based on the object of mentioning a method for suppressing an acoustic noise caused by noise, which allows the use of an adaptive filter, and at the same time has the highest possible sound quality in the output signal.

The above object is achieved by a method for suppressing a noise in an acoustic system, wherein the acoustic system comprises at least one microphone and at least one speaker, wherein the at least one microphone generates an input signal and wherein the at least one loudspeaker generates an acoustic signal, which partially feeds back to the at least one microphone, wherein along a main signal path in response to the input signal, a first intermediate signal and from the first intermediate signal by a frequency distortion an output signal is formed, wherein the output signal is coupled out of the main signal path in a signal feedback path, wherein in the signal Feedback path from the output signal by decorrelation a second intermediate signal is formed, which is used as an input to an adaptive filter, which generates a compensation signal, and wherein the Kompensat ion signal is supplied to the input signal for compensation, wherein from the input signal and / or from the compensated input signal, a third intermediate signal is formed, which is used as an input to the adaptive filter, and wherein the output signal is supplied to the at least one speaker for playback. Advantageous and partly inventive in themselves inventive embodiments are set forth in the dependent claims and the description below.

In particular, the output signal can also be used as a further input variable for the adaptive filter, wherein the second intermediate signal and the third intermediate signal are used in the adaptive filter for determining filter coefficients, by means of which the output signal is filtered and the compensation signal is thereby generated.

The invention is based on the following considerations: A reduction in the step size p of an adaptive filter used would result in the filter diverging much more slowly in the case of a correlated input signal, so that undesired artifacts in the output signal could be reduced or become inaudible. The reduction of the step size could, for example, always take place when a correlated or tonal input signal is registered. However, a disadvantage of such a procedure is that any change in the acoustic feedback path while the correlated signal is being registered can not be tracked fast enough to avoid feedback noise, because of the reduced step size p limitations on the adaptability of the filter be placed. The step size is therefore always to be seen as a trade-off between sound quality and the ability to respond to changes in the acoustic feedback path.

Another way to solve the problems of an adaptive filter for a strongly correlated input signal is a possible decorrelation of the input signal (so-called "pre-whitening"). Since only correlated input signals cause matching problems in the adaptive filter, such decorrelation could first solve the problem. Such decorrelation is often implemented by a linear predictor. For a correlated input signal, a prediction is made for one or more future samples of the signal as a function of past observed samples of the signal. This prediction then becomes the actual input signal subtracted. The result of this subtraction is called the prediction error signal ("residual signal"). For example, a sinusoidal signal is completely deterministic and therefore perfectly predictable. In this case, for a corresponding prediction order, the residual signal would be zero.

wobei s(k) das Sample des Input-Signals für die Prädiktion zum Zeitpunkt k darstellt, a(i) den Filterkoeffizienten der Dekorrelierung bezeichnet, und P die Ordnung der Prädiktion. Das so erzeugte Prädiktions-Fehlersignal ist dabei im Allgemeinen komplexwertig.In the case of linear prediction, the prediction error signal may be written as $$r\left(k\right)=s\left(k\right)-{\displaystyle \underset{i=1}{\overset{P}{\Sigma }}s\left(k-i\right)a\left(i\right)}.$$

where s (k) represents the sample of the input signal for the prediction at time k, a (i) denotes the filter coefficient of decorrelation, and P the order of prediction. The prediction error signal thus generated is generally of complex value.

Interference caused by feedback also has significantly correlated signal components. If a decorrelation is applied to such a signal, the signal strength of the resulting prediction error signal is very low. For reuse in an adaptive filter, this would mean that the adaptive filter is not excited at the frequency of the noise generated by the feedback. Thus, the filter can not adapt to the acoustic feedback path at this frequency, whereby the noise remains until the acoustic feedback path changes.

wobei der Vektor a die Koeffizienten a(i) enthält, und die Matrix R und der Vektor r die Autokorrelationsmatrix bzw. den Autokorrelationsvektor bezeichnet. Beide Größen werden gebildet durch die Autokorrelationen $$\mathrm{r}\left(\mathrm{j}\right)=\mathrm{E}\left\{\mathrm{s}\left(\mathrm{k}\right)\mathrm{s}\left(\mathrm{k}-\mathrm{j}\right)\right\},$$

wobei für stationäre Signale der Erwartungswert E nur von der Zeitverschiebung j abhängt. Der Erwartungswert kann dabei z.B. durch rekursives Mitteln approximiert werden.For estimating the filter coefficients for decorrelation by means of linear prediction, various methods exist, for example the NLMS algorithm and the Lewinson-Durbin recursion. In the latter, the following matrix-valued equation is recursively solved: $$a=R^{-1}r.$$

wherein the vector a contains the coefficients a (i), and the matrix R and the vector r the autocorrelation matrix or the Autocorrelation vector called. Both sizes are formed by the autocorrelations $$\mathrm{r}\left(\mathrm{j}\right)=\mathrm{e}\left\{\mathrm{s}\left(\mathrm{k}\right)\mathrm{s}\left(\mathrm{k}-\mathrm{j}\right)\right\}.$$

for stationary signals, the expected value E depends only on the time shift j. The expected value can be approximated, for example, by recursive means.

For non-stationary signals, e.g. Speech, the autocorrelation values are time-dependent, and therefore preferred to be repeatedly determined. However, most non-stationary signals can be considered nearly stationary within a time window of a certain duration. The length of this time window depends on the degree to which the signal is not stationary. The adaptation speed of a filter or estimator, which calculates the autocorrelation values of an input signal, plays an important role in this case: the faster the estimator, the better it is not possible to track stationary signals, which improves the decorrelation of an input signal. In order to be able to treat a non-stationary signal for a decorrelation within a short time window as stationary, it requires the fastest possible estimator. This also applies to those decorrelations which use a different method. In the NLMS algorithm, for example, the adaptation speed and thus the ability to decorrelate nonstationary signals is regulated by the step size.

The problem that a correlated target signal for the adaptive filter for canceling out a feedback-related noise is preferably decorrelated beforehand, but that a decorrelation of the adaptive filter at the frequencies of the feedback-induced noise no longer stimulates, could now be avoided, that in a first step, such a noise is detected, and in response to such detection in a second step in this case, the decorrelation omitted. However, this has several practical disadvantages: First, such a detection in practice is always faulty. In particular, if the acoustical feedback stimulates a plurality of closely spaced frequencies, these may not be sufficiently suppressed due to insufficient spectral resolution in the detection. Moreover, such a procedure initially always requires at least some development of a noise due to the feedback in order to bypass the corresponding signal processing block of the decorrelation when it is detected. This means that an internal signal in the acoustic system is never completely free from feedback, but contains signal components of the noise up to the threshold value of the detection. However, this is undesirable for reasons of sound quality.

Another possibility could be to determine the filter coefficients for the decorrelation in another acoustic system, and to transmit these filter coefficients continuously between the participating acoustic systems for adaptation. This possibility would be given in particular in a binaural hearing aid system. The above idea would be based on the assumption that the sound signals from the environment recorded by the respective acoustic systems have a high degree of similarity, but that noise caused by feedback in a single system affects only the individual acoustic system. Since feedback noise at a given frequency will most likely only occur in one acoustic system, the filter coefficients for decorrelation determined in another acoustic system may be considered a good estimate for the decorrelation of a target signal in the feedback acoustic system are used. For this, however, the presence of a further acoustic system is first required, which is often not the case. Moreover, as a result of the transmission also a time delay the filter coefficients occur so that they are no longer up-to-date when receiving in the other acoustic system, or due to the spatial arrangement of the acoustic systems involved, the respective filter coefficients are not a sufficiently good estimate for the other system. This may, for example, in a binaural hearing aid system caused by the head of the user shading effects occur.

In contrast, the invention proposes to first subject an output signal of the acoustic system, which is to be fed into a signal feedback path, to frequency distortion, and then to decorrelate it. In particular, a time-dependent frequency distortion can be used here. Noise caused by feedback usually has a nearly perfectly sinusoidal signal. Due to the frequency distortion, this form is lost. If, for example, a time-dependent frequency shift is selected for the frequency distortion, the signals of the noise follow this frequency shift.

The autocorrelation values of frequency-distorted signals decrease as the time interval increases, so that the window of time during which the feedback-induced noise can be considered stationary is shortened. Thus, it is possible to implement a decorrelator such that it does not adapt to the interference signal of the feedback. The time window in which signals can be considered stationary, is preferably to be chosen so that the interference signal of the feedback is not considered stationary by the frequency distortion, the actually non-stationary signal components of a target signal already. Thus, the decorrelation is not adapted to the interference signal, but only to the signal components of the target signal, which are decorrelated. In the decorrelated signal, the non-stationary correlated signal components are removed, as they occur in the recording of spoken speech, but not caused by the feedback Signal components. The decorrelated signal is now supplied as an intermediate signal to the adaptive filter, which can generate a compensation signal based on the feedback caused by feedback, which is fed back to the main signal path for the suppression of the noise.

The input signal is favorably time-discretized, whereby a "least mean square" algorithm (LMS) is used as the adaptive filter. In this case, the output signal is preferably used as the reference signal, and the error signal of the LMS filter is formed by the difference between the input signal and the compensation signal. The specified method is particularly advantageous in the use of an LMS algorithm in the adaptive filter, since the frequency distortion of the output signal solves the divergence problems which occur when using an LMS algorithm for the adaptive filtering of feedback-related interference signals.

It proves to be further advantageous if the step size in the LMS algorithm is normalized via the second intermediate signal. This procedure is also called "normalized least mean square" (NLMS). Such normalization improves the convergence properties of the algorithm. The optimal filter coefficients are generally given by the solution of the filter equation by means of a Wiener filter. However, due to the static properties and the limited conversion time, this can usually not be used, which is why estimates are used for the filter coefficients given by the Wiener filter, the estimates converging ideally against the Wiener solution. In the case of an LMS algorithm for estimating the optimum filter coefficient in the sense of a Wiener filter, too great a step size p in the vicinity of the optimal solution may worsen the convergence, since a relatively coarse movement around the optimal solution takes place in the solution space due to the iteration steps. By normalizing the step size and thus by the transition to the NLMS, the motion is refined near the optimal filter coefficients, thereby preventing excessive removal of the optimum solution in the solution space in the individual iteration steps.

Conveniently, the frequency distortion to form the output signal from the first intermediate signal is achieved by a frequency shift. In particular, a time-dependent frequency shift is used. This provides the ability to tune the decorrelator's adaptation speed to the frequency offset, and thus effectively exclude the frequency-shifted signal components of the acoustic feedback noise from the decorrelation. However, frequency distortion may also be by phase modification, frequency transposition, or non-linear transformation. Also in this case, the adaptation speed of the decorrelator is preferably to be tuned to the respective degree of frequency distortion.

It proves to be further advantageous if the output signal for decoding the second intermediate signal is decorrelated by means of a linear prediction filter. The filter coefficients of the linear prediction filter are preferably to be determined by means of a Levinson-Durbin recursion or by means of an LMS or NLMS algorithm. The advantage of a linear prediction filter is that only linear systems of equations have to be solved, which limits the numerical complexity for the respective filter problem. In particular, the input signal or the compensated input signal can also be decorrelated by means of a linear prediction filter, and used to form the third intermediate signal, which is supplied as an input variable to the adaptive filter.

In this case, time-dependent autocorrelation values of the output signal and / or of a signal based on the input signal are preferred for the filter coefficients of the linear prediction filter Error signal used. In particular, the autocorrelation values can be used for a Levinson-Durbin algorithm. Taking into account the time dependence of the autocorrelation values allows the decorrelation to be adjusted to the degree of frequency distortion via the appropriate choice of a corresponding time window, after which the autocorrelation values are again determined.

Particularly preferably, the filter coefficients of, in particular, each linear prediction filter are adapted as a function of the decorrelation strength of the frequency distortion. This means in particular that the time window in which signals can be regarded as stationary depends on the decorrelation strength of the frequency distortion. In the case of a Levinson-Durbin algorithm, this can be done, for example, via a repeated adaptation of the autocorrelation values in the mentioned time intervals, from which the filter coefficients are to be determined again. In the case of an NLMS algorithm, the step size can instead be adjusted accordingly in the time intervals specified.

The described functional dependence of the time intervals or of the stationary time window can influence which signal components are still perceived as stationary by the decorrelator, so that the signal components of the interference signal affected by the frequency distortion are not decorrelated. A decorrelator, which has too short a "stationary time window", could also perceive signal components of a frequency-distorted originally monofrequency signal as stationary and therefore decorrelate it. This is circumvented by adapting the rate of adaptation of the decorrelation to the degree of frequency distortion, in particular to that of its own decorrelation strength. If, for example, a time-dependent frequency shift is selected, then this is preferably carried out more quickly than signals are considered to be stationary in the time window for the decorrelation.

In a further advantageous embodiment of the invention, the filter coefficients of the, in particular each linear prediction filter are adapted in dependence on a transfer function of a model of the acoustic system, which comprises the at least one microphone and at least one speaker reproducing the corrected output signal. In particular, the time intervals for the adaptation of the filter coefficients may additionally depend on the decorrelation strength of the frequency distortion. The transfer function may hereby contain the specific characteristics of the acoustic system, e.g. Gain values in individual sub-bands. In such a model, it is also possible, at least implicitly via coefficients of the transfer function, to enter into the probability that a feedback causes noise at a certain frequency. If a feedback excitation is very likely or above a predetermined probability limit, the decorrelation adaptation rate may be reduced to ensure that the frequency-distorted components of the original monofrequency noise are not considered stationary and decorrelated. If feedback is unlikely, the time window for the decorrelator adaptation can be shortened so that tonal signal components, e.g. generated by voice recording, are quickly detected, and decorrelated.

The invention further provides an acoustic system comprising at least one microphone for generating an input signal, at least one loudspeaker for reproducing an output signal, and a control unit which is adapted to generate a noise by feedback of the output signal reproduced via the at least one loudspeaker is caused in the input signal generated by the at least one microphone, suppress by the above-described method. In particular, the acoustic system is here as a hearing aid, and advantageous as a Hearing aid formed. The advantages stated for the method and its developments can be transferred analogously to the acoustic system.

FIG 1

in einem Blockdiagramm der Ablauf eines Verfahrens zur Unterdrückung eines Störgeräusches in einem akustischen System, und

FIG 2

in einem Blockdiagramm eine weitere Ausgestaltungsmöglichkeit des Verfahrens nach FIG 1.

An embodiment of the invention will be explained in more detail with reference to a drawing. Here are shown schematically in each case:

FIG. 1

in a block diagram of the sequence of a method for suppressing a noise in an acoustic system, and

FIG. 2

in a block diagram, a further embodiment of the method according to FIG. 1 ,

Corresponding parts and sizes are provided in all figures with the same reference numerals.

In FIG. 1 3 shows a schematic block diagram of the sequence of a method 1 for suppressing a noise g in an acoustic system 2. The acoustic system 2, which is given here by a hearing device 3, for example a hearing aid device, comprises a microphone 4 and a loudspeaker 6. The microphone signal m recorded by the microphone 4 is fed to a signal processing unit 10 in a main signal path 8 where it amplifies, among other things becomes. At the end of the main signal path 8, an output signal xs is output to the microphone 4, which generates an acoustic signal p from the output signal xs. A part of the acoustic signal p generated by the loudspeaker 6 is again recorded by the microphone 4 as feedback fb, and thus finds its way into the microphone signal m. By means of the feedback fb, signal components of the acoustic signal p in the microphone signal m are fed again to the signal processing unit 10 and further amplified there. The repeated amplification, playback and recording in a closed process produces noise g in the form of almost monofrequente whistling sounds. To suppress the noise g, the signal feedback path 16 is provided.

For the signal feedback path 16, the output signal x s is coupled out of the main signal path 8 and fed to a decorrelator 18. The decorrelator 18 is formed by a linear prediction filter 20.

In the main signal path 8, the signal processing unit 10 outputs a first intermediate signal x, which is converted by a frequency distortion 22 in the output signal xs. The frequency distortion 22, which is achieved in the present case by a frequency shift 23, has the consequence that the linear prediction filter 20 does not decorrelate the signal components corresponding to the noise g, but only signal components of a target signal. From the linear prediction filter 20, a second intermediate signal xw is output as an input to an adaptive filter 24. The adaptive filter 24 generates from the output signal xs a compensation signal c, which is subtracted from the microphone signal m to compensate for the noise g. As a result, the signal feedback path 16 is closed.

For the generation of the compensation signal c, the adaptive filter 24 is supplied with an additional intermediate signal ew as an input signal. This third intermediate signal ew is formed from the error signal e, which results from the microphone signal m compensated for the compensation signal, c. The error signal e is now likewise decorrelated by a linear prediction filter 26, and the decorrelated error signal ew is supplied as a second input variable to the adaptive filter 24. From the decorrelated error signal ew and the second intermediate signal xw, the coefficients h are calculated in a filter block 28 of the adaptive filter 24, from which a signal block 30 of the adaptive filter together with the output signal xs generates the compensation signal c.

The frequency shift 23 ensures that the linear prediction filter 20 does not decorrelate any signal components associated with the noise g, whereby the adaptive filter 24 would no longer compensate for these with the compensation signal c. The length of the stationary time window T of the linear prediction filters 20, 26, and thus their adaptation speed, is thereby controlled as a function of the frequency shift 23. A control unit 32 in the hearing aid 3 performs all specified method steps.

In FIG. 2 is in a block diagram a slight modification of the in FIG. 1 shown method 1 shown. Here, in the acoustic system 2, ie in particular in a hearing aid 3, for example in a hearing aid, the decorrelated error signal ew, which is supplied as an input to the adaptive filter, is formed from an input signal mw decorrelated in the linear prediction filter 26 and a decorrelated compensation signal cw. The decorrelated compensation signal cw is formed in the filter block 28 of the adaptive filter from the error signal ew decorrelated in the linear prediction filter 26 and the second intermediate signal xw, which is given by the output signal xs decorrelated in the linear prediction filter 20. The length of the stationary time window T of the linear prediction filters 20, 26, and thus their adaptation speed, is hereby determined by an adaptation control 34, in which the degree df of the frequency shift 23, the gain n of the signal processing unit 10 in individual sub-bands, and a non find more detailed transfer function of the acoustic system 2 input and used to determine the time window T. Likewise, a model of the acoustic feedback path fb determined by the filter coefficients h can also be used, so that the adaptation speed of the decorrelation in the linear prediction filters 20, 26 is also determined as a function of the feedback estimated by this model. The use of such an adjustment control 34 is thereby not on the in FIG. 2 illustrated form of the signal feedback path 16 is limited, but can in principle in various embodiments, in particular in. In FIG. 1 shown embodiment, find use.

Although the invention has been illustrated and described in detail by the preferred embodiment, the invention is not limited by this embodiment. Other variations can be deduced therefrom by those skilled in the art without departing from the scope of the invention.

Claims (8)

1. Method (1) for suppressing an interference noise (g) in an acoustic system (2),
   wherein the acoustic system (2) comprises at least one microphone (4) and at least one loudspeaker (6),
   wherein the at least one microphone (4) generates an input signal (m), and wherein the at least one loudspeaker (6) generates an acoustic signal (p) which partially feeds back to the at least one microphone (4),
   wherein a first intermediate signal (x) is formed along a primary signal path (8) as a function of the input signal (m), and an output signal (xs) is formed from the first intermediate signal (x) via a frequency distortion (22),
   wherein the output signal (xs) is coupled out from the primary signal path (8) into a signal feedback path (16),
   wherein a second intermediate signal (xw) is formed in the signal feedback path (16) from the output signal (xs) via a decorrelation (18) by means of a linear prediction filter (20), wherein the filter coefficients of the linear prediction filter (20) are adapted as a function of the decorrelation strength of the frequency distortion (22), wherein the second intermediate signal (xw) is used as an input value for an adaptive filter (24), which generates a compensation signal (c), and wherein the compensation signal (c) is fed to the input signal (m) for compensation,
   wherein a third intermediate signal (ew) is formed from the input signal (m) and/or from the compensated input signal (e), which is used as an input value for the adaptive filter, and
   wherein the output signal (xw) is fed to the at least one loudspeaker (4) for reproduction.
2. Method (1) according to Claim 1,
   wherein the input signal (m) time-discretized, and wherein a least mean square (LMS) algorithm is used as an adaptive filter.
3. Method (1) according to Claim 2,
   wherein the increment in the LMS algorithm is normalized over the second intermediate signal (xw) .
4. Method (1) according to one of the preceding claims,
   wherein the frequency distortion (22) for forming the output signal (xs) from the first intermediate signal (x) is achieved via a frequency shift (23).
5. Method (1) according to one of the preceding claims,
   wherein time-dependent autocorrelation values of the output signal (xs) and/or an error signal (e) based on the input signal (m) are used for the filter coefficients of the linear prediction filter (20).
6. Method (1) according to one of the preceding claims,
   wherein the filter coefficients of the linear prediction filter (20) are adapted as a function of a transfer function of a model of the acoustic system (2), which comprises the at least one microphone (4) and at least one loudspeaker (6) reproducing the corrected output signal (xs).
7. Acoustic system (2), comprising at least one microphone (4) for generating an input signal (m), at least one loudspeaker (6) for reproducing an output signal (xs), and a control unit (32) which is configured to suppress an interference noise (g) due to feedback of the output signal (xs), which is reproduced via the at least one loudspeaker (6), into the input signal (m) generated by at least one microphone (4), via a method (1) according to one of the preceding claims.
8. Acoustic system (2) according to Claim 7, which is designed as a hearing device (3), in particular as a hearing aid device.

Images