JP5296247B1 - Sound processing apparatus and feedback cancellation method - Google Patents

Abstract

A technique for stably functioning a feedback canceller without using complicated processing is provided.
A feedback canceller 20 applied to a sound processing device uses a sound signal after hearing aid (amplification) processing by a hearing aid processing unit 12 as a reference signal u (n) of an adaptive filter 22, and an adaptive filter coefficient w _R. (N) is used to form the adaptive filter output signal y (n). The sum of the input signal s (n) of the ambient sound input to the microphone 10 and the feedback signal b (n) from the feedback path 16 is set as a desired signal d (n) of the adaptive filter 22, and the adaptive filter output signal y ( n) and an error signal e (n), which is the difference between the desired signal d (n), is provided to the hearing aid processor 12 to suppress howling. The adaptive filter coefficient calculation unit 24 updates the adaptive filter coefficient w _R (n) using an update formula obtained by synthesizing two algorithms used for the NLMS adaptive filter and the SLMS adaptive filter.
[Selection] Figure 1

Description

  The present invention relates to a sound processing apparatus and a feedback cancellation method suitable for suppressing howling that occurs in a hearing aid or the like.

  Conventionally, related art (Patent Documents 1 to 3) using various adaptive filters is known for a feedback canceller (AFBC) that suppresses howling of a hearing aid or the like.

  For example, as a first prior art, a feedback canceller of a hearing aid device (hearing aid) includes two adaptive filters that perform a rapid adaptive operation and an adaptive filter that performs a low-speed adaptive operation, and a method of switching between these two adaptive filters by a user operation (See Patent Document 1). In this method, low-speed adaptation is started by default, and when an undesirable noise is felt by the user due to a change in the sound environment, the user can switch to the rapid adaptation operation and immediately suppress howling.

  As a second prior art (Patent Document 2), a fixed filter that models an invariant feedback path unique to a hearing device and an adaptive filter that models a fluctuation part of the feedback path are provided. There are two methods using a slow fluctuation type and a sudden fluctuation type. In this method, which of the two adaptive filters is activated can be automatically switched by the control switch.

  In addition, as a third prior art (Patent Document 3), when a first adaptive filter and a second adaptive filter are provided in an internal circuit of a hearing aid, and the first adaptive filter functions as a feedback canceller, the adaptation is performed. There is a technique for optimizing the speed with a second adaptive filter. In this method, a normalized least mean square (Normalized LMS) algorithm with a slow adaptation speed is used for the first adaptive filter, but the output of the second adaptive filter does not affect the main signal processing (sound processing). Therefore, the adaptive filter coefficient can be updated using a faster sign least mean square (Signed LMS) algorithm.

Japanese Patent Laid-Open No. 6-189597 JP 2011-254468 A JP-T-2004-509543

  Both the first and second prior arts (Patent Documents 1 and 2) described above are techniques for switching the adaptation speed as necessary. In order for the feedback canceller to function properly, the timing for switching the adaptation speed is required. You must judge correctly. For this reason, when the adaptive speed is switched by a user operation as in the first prior art, there is a problem that the burden on the user is increased daily. Further, even when the adaptive speed is automatically switched as in the second prior art, there is a problem that a process for determining the switching timing is necessary in the control, and the processing load corresponding to that is increased.

  In the third prior art (Patent Document 3), only one type of application filter is used for the main signal path, and there is no need to determine the switching timing of the two application filters. However, since the adaptation speed of the main first adaptive filter (NLMS) is frequently changed by the second adaptive filter (SLMS), a complicated calculation process is required. Also, if the second adaptive filter lacks stability, the original feedback canceller may not function correctly.

  Therefore, an object of the present invention is to provide a technique capable of stably functioning a feedback canceller without using complicated processing.

In order to solve the above problems, the present invention employs the following solutions.
The sound processing device of the present invention amplifies the sound signal input inside by the sound processing means, and outputs the sound from the speaker based on the sound signal after the amplification processing, while the ambient sound collected by the microphone is Basically, the feedback component obtained by the sound output from the speaker reaching the microphone is input to the microphone as a feedback signal while the input signal is used. However, the howling caused by the feedback component is suppressed. Therefore, it has a function as a feedback canceller using an adaptive filter.

  In order to realize the function as a feedback canceller, the sound processing apparatus has a configuration as an adaptive filter, a feedback component removing unit, and an adaptive filter coefficient calculating unit. The feedback cancellation method of the present invention is realized by using the function of the sound processing device.

  In other words, the adaptive filter uses the amplified sound signal as a reference signal and convolves an adaptive filter coefficient with the reference signal to form an output signal in which a feedback component is estimated. Further, the feedback component removing means uses the sum of the input signal that is the input of the microphone and the feedback signal as a desired signal of the adaptive filter, and amplifies an error signal that is a difference between the desired signal and the output signal to the sound processing means. By inputting the sound signal before processing, the feedback component is removed from the sound signal after the amplification processing by the sound processing means. As a result, it is possible to suppress howling caused by the sound output from the speaker being fed back to the microphone.

  Further, the adaptive filter coefficient calculation means is a normalized LMS (Normalized Least Mean Square: hereinafter referred to as “NLMS”) that uses the reference signal and the error signal individually as the adaptive filter coefficient necessary for calculating the output signal by the adaptive filter. The algorithm and the signed LMS (Signed Least Mean Square: hereinafter referred to as “SLMS”) algorithm can be updated using one calculation formula obtained.

  As described above, the sound processing apparatus and the feedback cancellation method of the present invention apply a plurality of algorithms, NLMS and SLMS, when updating the adaptive filter coefficients, but two algorithms are used for calculating the actual adaptive filter coefficients. One calculation formula obtained by combining is used. That is, since a plurality of algorithms NLMS and SLMS are combined into one algorithm, no separate calculation process is required.

  In the present invention, in the process of updating the adaptive filter coefficient, for example, (1) an algorithm (SLMS) in which the initial adaptation speed is fast but the coefficient update is almost stopped when it converges to some extent, and (2) the adaptive speed. By combining both algorithms (NLMS) that are gentle but excellent in long-term stability, the advantages of each can be maximized. As a result, even if the desired characteristics such as “adaptive speed” and “stability” are in a trade-off relationship with only one adaptive filter of NLMS or SLMS, the advantages of each of the multiple types of adaptive filters can be obtained. By acting in a complementary manner, an ideal adaptive filter as a whole can be constructed.

  Further, according to the present invention, in the function as the feedback canceller, the function as the NLMS adaptive filter and the function as the SLMS adaptive filter can coexist. Therefore, for example, in the early stage of howling, when the howling is quickly suppressed by utilizing the function of the SLMS adaptive filter having a faster convergence speed, and the SLMS adaptive filter does not operate remarkably thereafter, as the NLMS adaptive filter, It is possible to maintain stability while maintaining high precision adaptation by function.

  As described above, according to the sound processing device and the feedback cancellation method of the present invention, the feedback canceller can function stably without using complicated processing.

It is a block diagram which shows the structure of a sound processing apparatus roughly. It is a figure which shows the characteristic acquired when an NLMS adaptive filter (large step size) is independently applied to a hearing aid model. It is a figure which shows the characteristic acquired when an NLMS adaptive filter (small step size) is independently applied to a hearing aid model. It is a figure which shows the characteristic acquired when a SLMS adaptive filter is applied independently to a hearing aid model. It is a figure which shows the characteristic of the adaptive filter using a peculiar algorithm. It is a block diagram showing the model assumed when two types of adaptive filters are mounted in a hearing aid.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

  FIG. 1 is a block diagram schematically showing a configuration of a sound processing apparatus according to an embodiment. The sound processing apparatus is suitable for use as a hearing aid 1, for example. Hereinafter, an embodiment of the hearing aid 1 will be described as an example. In addition, through the following description, a feedback cancellation method realized using the sound processing device will also be clarified.

  In the following formulas (1) to (12) and FIGS. 1 and 6, various symbols shown in italics represent variables. For convenience, various symbols are described in standard form in the text, but the corresponding symbols shown in italics in equations (1) to (12) and FIGS. 1 and 6 all represent variables.

  For example, the sound processing device as the hearing aid 1 includes a microphone 10, a hearing aid processing unit 12, and a speaker 14. A sound analog signal input to the microphone 10 is converted into a digital signal by an A / D converter (not shown). The hearing aid processor 12 amplifies (hears) the input digital signal (sound signal). The sound signal amplified by the hearing aid processing unit 12 is converted to an analog signal by a D / A converter (not shown) and output from the speaker 14 as sound. The speaker 14 provides output sound to the hearing aid user's ear (tympanic membrane).

[State model]
The basic operation of the hearing aid 1 has been described above. Considering, for example, a discrete-time state space model, the hearing aid processing unit 12 in the hearing aid 1 generally has a transfer function from the sound input to the output direction. It is represented by G (z) (amplification + delay). Further, since the sound output from the speaker 14 is fed back to the microphone 10, the feedback path 16 during this period is represented by a reverse transfer function F (z). Note that the transfer function F (z) of the feedback path 16 varies depending on the shape and structure of the hearing aid, the body structure and behavior of the wearer, the surrounding environment, and the like.

[Feedback canceller (AFBC)]
The hearing aid incorporates a feedback canceller 20 that suppresses howling using an adaptive filter 22. Further, the feedback canceller 20 has an adaptive filter coefficient calculation unit 24, and the adaptive filter coefficient calculation unit 24 updates the adaptive filter coefficient used in the adaptive filter 22 (discrete time update).

[State variable]
The state variables in the hearing aid model are an input signal s (n) corresponding to an input of ambient sound to the microphone 10 and a feedback signal b (n) corresponding to a feedback component from the feedback path 16. A sound (= s (n) + b (n)) obtained by adding the input signal s (n) and the feedback signal b (n) is input.

When the sound signal amplified by the hearing aid processor 12 is used as the reference signal u (n) of the adaptive filter 22, the feedback canceller 20 performs a convolution operation on the reference signal u (n) and the adaptive filter coefficient w _R (n). Then, the adaptive filter output signal y (n) is output.

On the other hand, the feedback canceller 20 assumes that the sound (= s (n) + b (n)) input to the microphone 10 is the desired signal d (n) of the adaptive filter 22, and then adds the desired signal d (n) by the adder 26. The adaptive filter output signal y (n) is subtracted, and the result is input to the hearing aid processor 12 as an error signal e (n). At this time, the adaptive filter 22 is expressed by a transfer function W _R (z) obtained by estimating the transfer function F (z) of the feedback path 16. As a result, the target of hearing aid (amplification) processing by the hearing aid processor 12 is the error signal e (n) from which the feedback component obtained by the sound of the speaker 14 reaching the microphone 10 is removed.

  A feature of this embodiment is an adaptive filter coefficient update algorithm in the adaptive filter coefficient calculation unit 24. Specifically, the adaptive filter coefficient is updated using a series of calculation formulas shown below.

[Adaptive filter output signal]
The adaptive filter output signal y (n) from the adaptive filter 22 is expressed by the following equation (1).
here,
w _R : Adaptive filter coefficient M: Number of taps of adaptive filter coefficient u: Reference signal. From the above equation (1), the adaptive filter output signal y (n) is obtained by time convolution of the adaptive filter coefficient and the reference signal.

[Error signal]
In the adder 26, the error signal e (n) is obtained by the following equation (2).

[Adaptive filter coefficient]
The adaptive filter coefficient calculation unit 24 calculates the updated adaptive filter coefficient w _R (k, n + 1) by the following equation (3).
here,
μ _N : Step size “￣u ² ” of normalized LMS (NLMS): Time average value of square of reference signal u μ _S : Step size “sign {}” of signed LMS (SLMS): sign function.

[Features of the algorithm]
The characteristic of the above equation (3) is that two types of algorithms, NLMS adaptive filter and SLMS adaptive filter, are combined into one update equation. That is, up to the second term on the right side of Equation (3) corresponds to the algorithm of the NLMS adaptive filter, and the third term on the right side corresponds to the algorithm of the SLMS adaptive filter. This will be described in more detail below.

[NLMS adaptive filter]
In general, the adaptive filter coefficient update formula of the NLMS adaptive filter is expressed by the following formula (4).

[SLMS adaptive filter]
In general, the adaptive filter coefficient update formula of the SLMS adaptive filter is expressed by the following formula (5).

  Note that the two equations (4) and (5) listed above are algorithms that can be realized when both are applied to the model of FIG. Therefore, when Equation (4) is applied alone to the adaptive filter 22 in the hearing aid model of FIG. 1, the feedback canceller 20 represents the characteristic characteristic of the NLMS adaptive filter. On the other hand, when Equation (5) is applied to the adaptive filter 22 in the hearing aid model of FIG. 1, the feedback canceller 20 represents the characteristic characteristic of the SLMS adaptive filter.

[Inherent characteristics of NLMS adaptive filter]
As shown in the above equation (4), the NLMS adaptive filter is an algorithm that divides the step size by the average power of the reference signal u (n). For this reason, the NLMS adaptive filter has a characteristic that the convergence speed of the adaptive filter does not depend on the amplitude of the reference signal u (n).

  Therefore, the NLMS adaptive filter not only suppresses howling because it continues to estimate the transfer function F (z) of the feedback path 16 even if the reference signal u (n) becomes extremely small after suppressing howling. Residual howling (so-called ringing phenomenon) caused by the feedback path 16 can also be suppressed.

The NLMS adaptive filter, by adjusting the step size mu _N (magnitude adjustment), it is possible to adjust the convergence rate (howling suppression speed). However, if the convergence is set too fast, the update of the filter coefficient becomes unstable and oscillation tends to occur. In particular, entrainment is likely to occur when the update of the adaptive filter coefficient becomes unstable when a signal with high purity (high periodicity) is input. Entrainment means that in a feedback canceller using an adaptive filter in a hearing aid, this feedback canceller malfunctions and distorts an input signal close to a sine wave. In order to suppress such a phenomenon, it is necessary to minimize the step size mu _N, too much small, rather the time to suppress the howling longer.

2 and 3 are diagrams showing characteristics obtained when the NLMS adaptive filter is applied alone to the hearing aid model. The difference in Figure 2 and Figure 3 is due to the difference of the step size mu _N of NLMS adaptive filter, if relatively large step size becomes the characteristic of FIG. 2, when a relatively small step size 3 It becomes the characteristic. At this time, (A) in FIG. 2 and (A) in FIG. 3 represent temporal changes in the sound pressure level from the time of howling. Further, (B) in FIG. 2 and (B) in FIG. 3 represent time changes in misalignment of adaptive filter coefficients.

[Definition of misalignment]
Here, the misalignment (M (w (n)) indicates how close the correct value f (n) of the transfer function F (z) estimated by the adaptive filter coefficient w (n) is at a certain time n. That is, misalignment means that the smaller the value, the smaller the estimation error, and it is defined by the following equation (6).
Here, || || represents a norm. When w (n) can completely estimate the correct value f (n), the misalignment is −∞ [dB] from the above equation (6).
In the above formula (6), the symbols “w” and “f” in bold letters mean vectors, but in this paragraph, a standard is used for convenience (the same applies below).

[When step size is large]
As shown in FIG. 2 (A), the by relatively large step size mu _N of NLMS adaptive filters, it is possible to more rapidly suppressing feedback. However, as shown in FIG. 2B, it can be seen that the misalignment continues to fluctuate even after the howling is suppressed, and the operation is still unstable. Therefore, if a high-pure sound is input during this period, the oscillation phenomenon is likely to occur as described above.

[When the step size is small]
In contrast, when a relatively small step size mu _N of NLMS adaptive filter, as shown in FIG. 3 (A), the it takes a long time to suppress the howling. Instead, as shown in FIG. 3B, it can be seen that the operation is extremely stable because the misalignment shows only a decreasing tendency after the howling is suppressed. Therefore, even if a pure tone sound is input during this period, the oscillation phenomenon hardly occurs.

[Inherent characteristics of SLMS adaptive filter]
Next, as shown in the above equation (5), the SLMS adaptive filter is an algorithm that encodes the reference signal u (n) into binary values of “+1” and “−1”. That is, “+1” is assigned if the amplitude of the reference signal u (n) is on the plus side or on the minus side, and “−1” is assigned if the amplitude is on the minus side. As a result, the binarized signal assigned to “+1” or “+1” is used as it is for the calculation of the adaptive filter coefficient. For this reason, the SLMS adaptive filter has an advantage that the amount of calculation can be relatively reduced.

The SLMS adaptive filter, by selecting the step size mu _S appropriately, while faster speeds of up to suppress howling in accordance later suppression update of the adaptive filter coefficients is reduced, the update of the adaptive filter coefficients It is possible to make a setting that substantially stops. Therefore, it is possible to realize an algorithm in which an oscillation phenomenon (entrainment) is unlikely to occur by stopping the update of the adaptive filter coefficient while suppressing howling as quickly as possible.

  However, a major problem with the SLMS adaptive filter is that once the howling stops, the updating of the adaptive filter coefficient almost stops there, so that the residual howling (ringing) phenomenon cannot be suppressed.

  FIG. 4 is a diagram showing characteristics obtained when an SLMS adaptive filter is applied alone to a hearing aid model. Similarly, (A) in FIG. 4 represents a time change in sound pressure level from the time of occurrence of howling, and (B) in FIG. 4 represents a time change in misalignment of adaptive filter coefficients.

  As shown in FIG. 4A, it can be seen that the SLMS adaptive filter can suppress howling very rapidly. However, as shown in FIG. 4B, after the howling is suppressed, the update of the adaptive filter coefficient is almost stopped, and no further operation is performed. Therefore, as described above, it is not suitable for suppressing the residual howling (ringing) phenomenon.

[Characteristic Evaluation of Adaptive Filter of this Embodiment]
FIG. 5 is a diagram illustrating characteristics of an adaptive filter using an algorithm unique to the present embodiment. Below, the usefulness of this embodiment is evaluated through the comparison with each characteristic at the time of applying NLMS adaptive filter and SLMS adaptive filter individually, respectively.

[Adaptive speed evaluation]
As shown in FIG. 5A, by using the adaptive filter of the present embodiment, it is possible to suppress howling within a short time (in this example, shorter than 1 sec) from the occurrence. I understand. This adaptive speed is faster than when the step size of the NLMS adaptive filter is relatively large ((A) in FIG. 2), and also when the SLMS adaptive filter is used alone ((A) in FIG. 3). Comparable.

[Stability evaluation]
(B) in FIG. 5 represents the time change of the misalignment of the adaptive filter coefficient obtained by the algorithm of the present embodiment. The change in misalignment shown here is due to (depends on) the entire algorithm specific to this embodiment. In this case, the calculated value of the filter coefficient rapidly approaches the correct value from the time of occurrence of howling, and the filter coefficient is continuously updated even after suppressing howling. You can see that you can keep approaching.

  That is, in the algorithm specific to the present embodiment, the portion corresponding to the SLMS adaptive filter is specialized for quick suppression of howling, and hardly moves in other states. On the other hand, the portion corresponding to the NLMS adaptive filter continues to operate to accurately estimate the feedback component even after once suppressing howling. For this reason, it is preferable to set the NLMS to a relatively slow update speed (small step size). Accordingly, it is possible to stably maintain a state in which an oscillation phenomenon (ringing) hardly occurs and an entrainment hardly occurs while suppressing howling quickly.

  In this embodiment, even if two types of adaptive filters, NLMS and SLMS, are used as a concept, one new adaptive filter 22 is actually constructed in the model. Therefore, a complicated process of selecting one of the adaptive filter coefficients depending on the case is not required.

  Furthermore, since the convergence speed of the portion corresponding to the NLMS adaptive filter having a relatively large amount of calculation may be slow, the frequency of calculating and updating the adaptive filter coefficient can be greatly reduced with respect to the SLMS adaptive filter. As a result, the processing load on the processor can be reduced in actual hardware implementation, and the power consumption can be kept low.

〔Comparison〕
From the above, the usefulness of the algorithm specific to the present embodiment has already been clarified, but the inventors of the present invention have made further pursuits, and the algorithm specific to the present embodiment is merely a set of two types of adaptive filters. It makes clear that it is not a thing. Hereinafter, this point will be referred to.

  FIG. 6 is a block diagram showing a model assumed when two types of adaptive filters are simply implemented in a hearing aid (reference numeral omitted). This model simply assumes that two NLMS adaptive filters and SLMS adaptive filters are operated simultaneously (in parallel).

  In this case, the hearing aid is provided with two independent feedback cancellers 40 and 50, and different types of adaptive filters 42 and 52 are arranged in each of them. Here, it is assumed that the NLMS adaptive filter 42 is used for one feedback canceller 40 and the SLMS adaptive filter 52 is used for the other feedback canceller 50. Therefore, each of the feedback cancellers 40 and 50 requires an NLMS adaptive filter coefficient calculation unit 44 and an SLMS adaptive filter coefficient calculation unit 54 according to the algorithm of the adaptive filters 42 and 52.

Further, since two independent feedback cancellers 40 and 50 are incorporated in one hearing aid system, one of them is incorporated in the other loop (so-called “nested” state). In this case, two adders 56 and 46 (shown in the order of directions) are placed between the microphone 10 and the hearing aid processor 12, and the desired signal d (n) and the adaptive filter output signal y are included in the front adder 56. Although the error signal _e S and _S (n) (n) is calculated, the next addition unit 46, the error signal _e n adaptive filter output signal _y n (n) and the error signal _e S (n) ( n) will be computed.

[Calculation formula for adaptive filter update]
In the model of FIG. 6, the following series of calculation formulas are required for updating the adaptive filter.

[SLMS adaptive filter output signal]
The adaptive filter output signal y _S (n) of the SLMS adaptive filter 52 is calculated (convolution operation) by the following equation (7).
here,
w _S : SLMS adaptive filter coefficient M: The number of taps of the adaptive filter coefficient.

[Error signal]
In the front addition unit 56, the error signal e _S (n) is calculated by the following equation (8).

[SLMS adaptive filter coefficient]
The SLMS adaptive filter coefficient calculation unit 54 calculates the updated adaptive filter coefficient w _S (k, n + 1) by the following equation (9).

[NLMS adaptive filter output signal]
In the inner loop, the adaptive filter output signal y _N (n) of the NLMS adaptive filter 42 is calculated (convolution operation) by the following equation (10).
here,
w _N : NLMS adaptive filter coefficient M: the number of taps of the adaptive filter coefficient.

[Error signal]
In the rear addition unit 46, the error signal e _N (n) is calculated by the following equation (11).

[NLMS adaptive filter coefficient]
Then, the NLMS adaptive filter coefficient calculation unit 44 calculates the updated adaptive filter coefficient w _N (k, n + 1) by the following equation (12).

[Comparison with this embodiment]
Although the simple model of FIG. 6 can obtain the same or approximate performance as the present embodiment, it is necessary to separately calculate two adaptive filter coefficients in order to operate the two types of adaptive filters 42 and 52 in parallel. The amount of calculation has increased accordingly.
On the other hand, this embodiment has an advantage that it can operate optimally with only one adaptive filter 22 and that the amount of calculation can be overwhelmingly less than that of the model of FIG.

  The present invention can be implemented with various modifications without being limited to the above-described embodiment. In one embodiment, two types of algorithms, a general NLMS adaptive filter and an SLMS adaptive filter, are combined into one to form a unique algorithm, but other types of algorithms may be combined. Further, the adaptive filter coefficients may be calculated by synthesizing a plurality of types of adaptive filter algorithms that are more finely divided.

  For example, the adaptive filter coefficient can be updated by arbitrarily combining a plurality of types of adaptive filters whose response characteristics such as misalignment tend to be complementary, and combining the respective algorithms. There is a wide range of arbitrary characteristics in combining adaptive filters of any nature, and there are no particular restrictions on the implementation of the present invention.

  The present invention can be implemented not only as an embodiment as a hearing aid but also as various acoustic devices such as a telephone, an acoustic device with a microphone mixing function, a loudspeaker, a recording device with a monitor function, and a broadcasting device. Further, the feedback cancellation method of the present invention can be realized by various acoustic devices.

10 Microphone 12 Hearing Aid Processing Unit (Sound Processing Means)
14 Speaker (Earphone)
20 feedback canceller 22 adaptive filter 24 adaptive filter coefficient calculation unit (adaptive filter coefficient calculation means)
26 Adder (feedback component removal means)

Claims (2)

Sound processing means for amplifying the input sound signal;
A speaker that outputs sound based on the sound signal after amplification processing by the sound processing means;
While using the collected ambient sound as an input signal, a microphone that inputs a feedback component obtained by the sound output from the speaker reaching the microphone together as a feedback signal;
An adaptive filter that uses the sound signal amplified by the sound processing means as a reference signal, convolves an adaptive filter coefficient with the reference signal to form an output signal that estimates the feedback component;
The sum of the input signal that is the input of the microphone and the feedback signal is used as a desired signal of the adaptive filter, and an error signal that is the difference between the desired signal and the output signal is processed by the sound processing means before amplification processing. Feedback component removing means for removing the feedback component from the sound signal after amplification processing by the sound processing means,
A normalized LMS (Normalized Least Mean Square) algorithm and a signed LMS (Signed Least Mean Square) using the reference signal and the error signal individually as the adaptive filter coefficients necessary for the calculation of the output signal by the adaptive filter. Adaptive filter coefficient calculation means for updating using one calculation formula obtained by combining the algorithm ,
In the discrete-time system, the desired signal is d (n), the output signal is y (n), the error signal is e (n), the reference signal is u (nk), and the adaptation used in the adaptive filter. When the filter coefficient is w _R (k, n) and the number of taps of the adaptive filter coefficient is M, the output signal y (n) is expressed by the following equation (1).
And the error signal e (n) is obtained by the following equation (2):
When obtained by
When the step size of the normalized LMS algorithm is μ _N and the step size of the signed LMS algorithm is μ _S , the adaptive filter coefficient w _R (k, n + 1) updated by the adaptive filter coefficient calculation means is Formula (3)
A sound processing apparatus calculated by the following . While amplifying the sound signal input by the sound processing unit inside the sound processing device, and outputting the sound from the speaker based on the sound signal after the amplification processing, while using the ambient sound collected by the microphone as the input signal, In the process in which the feedback component obtained by the sound output from the speaker reaching the microphone is input to the microphone together as a feedback signal,
  The sound signal after the amplification processing by the sound processing unit is used as a reference signal for an adaptive filter inside the sound processing device, and an output signal obtained by convolution of the adaptive filter coefficient by the adaptive filter with respect to this reference signal and estimating the feedback component is obtained. Forming,
  The sum of the input signal that is an input of the microphone and the feedback signal is set as a desired signal of the adaptive filter in the sound processing device, and an error signal that is a difference between the desired signal and the output signal is set as the sound processing unit. By inputting as a sound signal before the amplification processing for, while removing the feedback component from the sound signal after the amplification processing by the sound processing unit,
  A normalized LMS (Normalized Least Mean Square) algorithm and a signed LMS (Signed Least Mean Square) using the reference signal and the error signal individually as the adaptive filter coefficients necessary for the calculation of the output signal by the adaptive filter. In updating using one calculation formula obtained by combining the algorithm,
  In the discrete-time system, the desired signal is d (n), the output signal is y (n), the error signal is e (n), the reference signal is u (nk), and the adaptation used in the adaptive filter. Filter coefficient w _R (K, n), where the number of taps of the adaptive filter coefficient is M, the output signal y (n) is expressed by the following equation (1).
And the error signal e (n) is obtained by the following equation (2):
When obtained by
  The step size of the normalized LMS algorithm is μ _N , The step size of the signed LMS algorithm is μ _S The adaptive filter coefficient w after being updated by the adaptive filter coefficient calculating means. _R (K, n + 1) is the following formula (3)
A feedback cancellation method characterized by being calculated by: