Feedback canceller, audio device equipped with same, and feedback cancellation method

Abstract

To provide a technique for surely suppressing howling while improving estimation accuracy of an adaptive filter.SOLUTION: A coefficient update unit 13 updates a coefficient of an adaptive filter 11 based on a coefficient W0 of an initial transfer function in addition to an error signal e(n) and a signal u(n) after delay processing using an NLMS algorithm. Specifically, in a case where a total sum of a current coefficient of the adaptive filter 11 and a differential of the coefficient W0 of the initial transfer function for each tap is smaller than a predetermined value (=the current coefficient is close to the coefficient W0), the coefficient is updated while taking the coefficient W0 into account (S140:Yes→S150) and in a case where equal to or greater than the predetermined value (=the current coefficient is not close to the coefficient W0), the coefficient is updated without taking the coefficient W0 into account (S140:No→S160). In such a configuration, even in a case where a level of an input signal is low or a closed loop gain is lowered in a quiet environment, updating of the adaptive filter is not delayed. Thus, estimation accuracy is improved and howling can be surely suppressed.SELECTED DRAWING: Figure 2

Description

[Technical field]

[0001] The present invention relates to a feedback canceller, an audio device including the same, and a feedback cancellation method. [Background technology]

[0002] Hearing aids are always faced with the problem of acoustic feedback accompanied by howling because the microphone and earphone are close to each other and high-gain amplification is performed. In order to suppress howling, it is essential to construct an adaptive filter that estimates the impulse response. However, the impulse response is different for each hearing aid wearer and time-varying. Furthermore, it is necessary to achieve low-delay, high-quality sound output and howling suppression at the same time under strict constraints on computing power. Therefore, the problem of acoustic feedback in hearing aids is difficult, and various methods have been tried to solve it. For example, a method of constructing an adaptive filter using the LMS algorithm as an adaptive algorithm is generally known. In addition, a technology that partially improves this method has been disclosed (for example, see Non-Patent Document 1). [Prior art documents] [Non-patent literature]

[0003] [Non-Patent Document 1] James M. Kates, “Constrained adaptation for feedback cancellation in hearing aids,” The Journal of the Acoustical Society of America, Vol. 106, No. 2, p. 1010-1019, August 1999. Summary of the Invention [Problem to be solved by the invention]

[0004] A major issue in howling suppression processing in hearing aids is entrainment (a phenomenon in which the input signal is distorted and produces abnormal sounds) and temporary howling, which occur due to a decrease in the estimation accuracy of the adaptive filter when a periodic signal is input. Factors that decrease the estimation accuracy of the adaptive filter include a decrease in the closed loop gain of the hearing aid and a decrease in the level of the input signal. Possible cases in which the closed loop gain decreases include when the volume of the hearing aid is low, or when the level of the sound fed back to the microphone decreases due to a process that suppresses noise and reverberation (hereinafter referred to as "noise suppression processing").

[0005] Here, the conditions for howling oscillation will be described with reference to FIG. Figure 11 (A) shows the basic configuration of a hearing aid capable of performing feedback suppression processing. In the configuration diagram, "F(z)" is the feedback transfer function, "W(z)" is the transfer function estimated by the adaptive filter, and "G(z)" is the hearing aid processing including amplification processing and noise suppression processing. FIG. 11(B) shows the closed loop transfer function C(z) of this hearing aid. FIG. 11C shows the oscillation conditions of howling when the howling suppression process is executed in this hearing aid. i " is the i-th frequency that satisfies the three conditional expressions. Also, "D(ω i ) is the differential transfer function obtained by subtracting the estimated feedback transfer function from the feedback transfer function. In other words, the howling oscillation condition is met at a specific frequency where the open-loop transfer function is 1 or more and the phase is aligned.

[0006] In conventional adaptive filters using the LMS algorithm, when the closed loop gain is reduced by noise suppression processing or when the input signal level is low, the update of the adaptive filter coefficients may be delayed. As a result, the estimation accuracy of the adaptive filter decreases, and the difference between "F(z)" and "W(z)" increases, making it more likely that the oscillation condition described above will be met. Furthermore, when a sudden sound is input during noise suppression processing, the noise suppression processing is released and the gain of the hearing aid rises sharply, causing "G(ω i ) will be larger, so the possibility of satisfying the above oscillation conditions will increase.

[0007] On the other hand, in the technology described in the above non-patent document, the impulse response (hereinafter referred to as the "coefficient of the initial transfer function") in the initial wearing state of the hearing aid is measured, and a configuration is adopted to prevent the coefficient of the adaptive filter from deviating significantly from the coefficient of the initial transfer function. Therefore, with this technology, it is considered that the decrease in estimation accuracy can be suppressed within a certain range. However, with this technology, if the impulse response changes significantly from the initial wearing state, there is a risk that howling will not stop. It is desirable that the adaptive filter not only has high estimation accuracy, but also can suppress howling even if the impulse response changes significantly from the initial wearing state.

[0008] Therefore, an object of the present invention is to provide a technique for reliably suppressing howling while improving the estimation accuracy of an adaptive filter. [Means for solving the problem]

[0009] In order to solve the above problems, the present invention employs the following feedback canceller, an audio device including the same, and a feedback cancellation method. Note that the following wording in parentheses is merely an example, and the present invention is not limited thereto.

[0010] That is, the feedback canceller of the present invention comprises an adaptive filter that estimates an impulse response from a sound output means to a sound input means, and outputs a signal equivalent to an estimated value of a feedback component based on a reference signal input to the adaptive filter; a subtraction unit that outputs an error signal obtained by subtracting the output signal of the adaptive filter from the output signal of the sound input means; a memory unit that stores coefficients of an initial transfer function based on a feedback impulse response specific to the usage environment, which is measured in advance using the sound output means and the sound input means; and a coefficient update unit that calculates an update amount for the adaptive filter coefficients based on the reference signal and the error signal, and if the difference between the current coefficients of the adaptive filter and the coefficients of the initial transfer function is smaller than a predetermined value, updates the coefficients of the adaptive filter taking the coefficients of the initial transfer function into account in the update amount, and if the difference is equal to or greater than the predetermined value, updates the coefficients of the adaptive filter without taking the coefficients of the initial transfer function into account.

[0011] In a conventional feedback canceller using the LMS algorithm, when the input signal level is low in a quiet environment or when the closed-loop gain is low, updating of the adaptive filter coefficients is delayed and a state in which the estimation error is large continues. In such a state, when a sudden sound is input, temporary howling may occur.

[0012] In contrast, in the feedback canceller of the above aspect, the coefficients of the initial transfer function (impulse response in the initial state) specific to the usage environment measured in advance are used to update the coefficients of the adaptive filter. Therefore, according to the feedback canceller of the above aspect, even when the level of the input signal is low in a quiet environment or the closed loop gain is low, the update of the adaptive filter is not delayed, so that the estimation error can be reduced to improve the estimation accuracy, and howling can be reliably suppressed.

[0013] Preferably, the feedback canceller of the above aspect further comprises a band limiting unit that limits a frequency band of at least one of the reference signal and the error signal, and the coefficient updating unit calculates an update amount of the adaptive filter coefficient based on an output signal of the band limiting unit. Alternatively, the feedback canceller of the above aspect further comprises a whitening unit that whitens at least one of the reference signal and the error signal, and the coefficient updating unit calculates an update amount of the adaptive filter coefficient based on an output signal of the whitening unit.

[0014] According to the feedback canceller of these aspects, it is possible to reduce the estimation error and improve the estimation accuracy, and also to suppress howling more reliably.

[0015] More preferably, in the feedback of any of the above aspects, when the difference is smaller than a predetermined value, the coefficient update unit updates the current coefficient by subtracting from the update amount a value obtained by multiplying the difference between the current coefficient of the adaptive filter and the coefficient of the initial transfer function by a predetermined constant, and when the difference is equal to or greater than the predetermined value, updates the current coefficient of the adaptive filter by adding the update amount.

[0016] According to the feedback canceller of this aspect, when the difference is smaller than a predetermined value, a constant based on a predetermined value determined based on the results of verification on an actual device or the results of a simulation is used to update the coefficients of the adaptive filter. This makes it possible to fine-tune the value to be added to the current coefficients, thereby making it possible to adjust the influence of the coefficients of the initial transfer function to suit the conditions of use.

[0017] In addition, the acoustic equipment of the present invention includes at least a feedback canceller according to any one of the above aspects, a sound input means for converting an input sound into an electrical signal and outputting it to the subtraction section, and a sound output means for converting an electrical signal based on an error signal into sound and outputting it.

[0018] According to this aspect of the audio device, an electrical signal corresponding to sound input to a sound input means (e.g., a microphone) is output to a feedback canceller (specifically, a subtraction unit), and a sound corresponding to an electrical signal based on the error signal from which the feedback component has been removed by the feedback canceller (the electrical signal after various acoustic processes including gain amplification have been applied to the error signal) is output from a sound output means (e.g., earphones or speakers), thereby reliably suppressing howling.

[0019] In addition to the steps executed by each component of the feedback canceller described above, the feedback cancellation method of the present invention includes a measurement step of measuring the coefficient of an initial transfer function based on a feedback impulse response specific to the usage environment, for the impulse response from the sound output means to the sound input means (between the sound output point and input point).

[0020] Preferably, in the feedback cancellation method of this aspect, in the measurement step, the initial transfer function is measured multiple times or for a certain period of time, and the measurement is terminated if a difference between the measurement results is within a predetermined range, and the measurement is allowed to continue if the difference exceeds the predetermined range.

[0021] According to this aspect, if the difference between the measurement results exceeds a predetermined range, the measurement is continued. This makes it possible to prevent accidental events from affecting the measurement results and influencing the coefficients of the initial transfer function, thereby contributing to improving the accuracy of estimation in the estimation process.

[0022] More preferably, in the feedback cancellation method according to any one of the above aspects, in the measurement step, the measurement is terminated when the number of measurements performed exceeds a predetermined maximum number.

[0023] According to this aspect, the measurement is not continued beyond a predetermined maximum number of times, so that the burden on the subject of measurement (the burden on the user of the acoustic equipment and the usage environment, in particular the mental and time burden on the user (human)) can be kept to a minimum.

[0024] More preferably, the feedback cancellation method of any of the above aspects further includes a noise estimation step of estimating at least one of the level of the ambient noise and its influence prior to measurement in the measurement step, and in the measurement step, the intensity of the acoustic signal used in the measurement in the measurement step is adjusted depending on the estimation result in the noise estimation step.

[0025] According to this aspect, the intensity of the acoustic signal (probe noise) used in the measurement is adjusted according to the level of the ambient noise and its influence, so that it is possible to prevent the ambient noise from affecting the measurement results and influencing the coefficients of the initial transfer function, which can contribute to improving the accuracy of estimation in the estimation process. Effect of the Invention

[0026] As described above, according to the present invention, it is possible to reliably suppress howling while improving the estimation accuracy of the adaptive filter. [Brief description of the drawings]

[0027] [Figure 1] 1 is a block diagram showing the configuration of a hearing aid 100 including a feedback canceller 10 of a first embodiment. [Diagram 2] FIG. 4 is a diagram illustrating an example of a procedure of a coefficient update process in the first embodiment. [Diagram 3] FIG. 13 is a block diagram showing a configuration related to measurement of a coefficient W0 of an initial transfer function. [Figure 4] FIG. 1 is a block diagram showing the configuration of a hearing aid 100' including a feedback canceller 10' as a comparative example. [Diagram 5] FIG. 11 is a diagram illustrating an example of a procedure for coefficient update processing in the comparative example. [Figure 6] FIG. 11 is a block diagram showing the configuration of a hearing aid 200 including a feedback canceller 20 of a second embodiment. [Figure 7] FIG. 11 is a diagram illustrating an example of a procedure of a coefficient update process in the second embodiment. [Figure 8]FIG. 11 is a block diagram showing the configuration of a hearing aid 300 including a feedback canceller 30 of a third embodiment. [Figure 9] FIG. 13 is a diagram illustrating an example of a procedure of a coefficient update process in the third embodiment. [Figure 10] 11 is a diagram showing a comparison of simulation results of an adaptive filter between an embodiment and a comparative example. FIG. [Figure 11] FIG. 2 is a diagram for explaining oscillation conditions of howling. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0028] Hereinafter, embodiments of the present invention will be described with reference to the drawings. Note that the following embodiments are preferred examples, and the present invention is not limited to these examples.

[0029] [First embodiment] FIG. 1 is a block diagram showing the configuration of a hearing aid 100 including a feedback canceller 10 according to the first embodiment.

[0030] 1, the feedback canceller 10 is composed of, for example, an adaptive filter 11, a subtraction unit 12, a coefficient update unit 13, an initial value storage unit 14, a delay processing unit 15, etc. The feedback canceller 10 employs the NLMS algorithm as an adaptive algorithm. The hearing aid 100 is composed of, for example, the feedback canceller 10, a microphone 50, a hearing aid processing unit 60, an amplifier unit 70, an earphone 80, etc., and is driven by a battery (not shown). Of the components of the hearing aid 100, the components other than the microphone 50 and the earphone 80 can be implemented by signal processing using, for example, a DSP (Digital Signal Processor).

[0031] In addition, since the occurrence of acoustic feedback requires the presence of both a means for outputting sound and a means for inputting sound, it is also possible to regard the microphone 50 and the earphone 80 as part of the feedback canceller 10.

[0032] In addition to a sound (hereinafter referred to as "external sound") that is a mixture of desired sounds such as voice and environmental sounds, the microphone 50 also receives a sound (hereinafter referred to as "feedback sound") that is a sound x(n) output from the earphone 80 that leaks from the ear canal and is fed back to the microphone 50 via a feedback transfer function F(z). The microphone 50 converts a composite sound of the input external sound s(n) and the feedback sound f(n) into an electrical signal, and outputs a signal d(n) that has passed through the transfer characteristics unique to the microphone 50.

[0033] The adaptive filter 11 adaptively estimates a transfer function W(z) corresponding to the feedback transfer function F(z) using coefficients updated by a coefficient update unit 13 described later, and outputs a signal y(n) obtained by passing a signal u(n) input to the adaptive filter 11 through the transfer function W(z). In other words, the signal y(n) corresponds to an estimate of the component of the feedback sound fed back from the earphone 80 to the microphone 50. The subtraction unit 12 subtracts the output signal y(n) of the adaptive filter 11 from the output signal d(n) of the microphone 50, and outputs the error signal e(n) obtained by this to the hearing aid processing unit 60 and the coefficient update unit 13.

[0034] The hearing aid processor 60 performs hearing aid processing G(z) on the error signal e(n), and outputs a post-hearing aid processing signal b(n) to the amplifier 70 and a delay processor 15, which will be described later. The amplifier 70 amplifies the post-hearing aid processing signal b(n) with a predetermined gain, and outputs the signal that has passed through a transfer characteristic specific to the amplifier 70 to the earphone 80. The earphone 80 converts the output signal of the amplifier 70 into a sound x(n) that has passed through a transfer characteristic specific to the earphone 80, and outputs the sound. Then, a portion of the sound output from the earphone 80 is fed back to the microphone 50, as described above.

[0035] Meanwhile, delay processing unit 15 outputs the post-hearing-aid processing signal with a predetermined delay in order to compensate for delays occurring in hearing aid 100, i.e., delays associated with hearing aid processing and delays when feedback sound reaches microphone 50 from earphone 80. Specifically, delay processing unit 15 outputs the post-hearing-aid processing signal that was input a predetermined time ago to adaptive filter 11 and coefficient update unit 13 as post-delay processing signal u(n).

[0036] Then, the coefficient update unit 13 uses the NLMS algorithm to update the coefficients of the adaptive filter 11 so as to minimize the error signal. Specifically, the coefficient update unit 13 updates the coefficients of the adaptive filter 11 based on the input error signal e(n) and the delayed signal u(n), as well as on the coefficient W0 of the initial transfer function measured at the time of initial fitting of the hearing aid, which is stored in advance in the initial value storage unit 14.

[0037] In the above example configuration, an amplifier unit 70 is provided, but instead, it is possible to configure the hearing aid processing unit 60 to perform gain amplification in addition to various hearing aid processes without providing the amplifier unit 70. In the following description, the signal input to the adaptive filter (the signal referenced by the adaptive filter) may be referred to as the "reference signal of the adaptive filter."

[0038] 2 is a diagram showing an example of the procedure of the coefficient update process in the first embodiment. Of these, (A) is a flowchart showing an example of the procedure of the coefficient update process executed by the coefficient update unit 13, and (B) is a specific algorithm of the example of the procedure. The following is an explanation of the example of the procedure.

[0039] Step S110: The coefficient update unit 13 calculates the sum ξ of the coefficients of all taps of the initial transfer function. Step S110 corresponds to the first line of the algorithm, and the sum ξ is calculated by the following formula.

[0040]

number

[0041] In the above formula, "N" is the tap length and "w k "(0)" is the coefficient of the k-th tap of the coefficient of the initial transfer function. "γ" is a constant that determines the degree of influence of the coefficient of the initial transfer function, more specifically, a constant used to adjust the true decision rate (the rate at which the decision formula holds) in the threshold decision described later, and is determined in advance based on the verification results of the actual device or the simulation results. The sum ξ may be calculated from the coefficient W0 of the initial transfer function that is stored in advance in the initial value storage unit 14, or the sum of the coefficients calculated in advance may be stored in the initial value storage unit 14 and read out and used when the step is executed. The measurement of the coefficient W0 of the initial transfer function will be described in detail later with reference to another drawing.

[0042] Step S120: The coefficient update unit 13 receives an input of a signal at the current time n. Specifically, the error signal e(n) and the delayed signal u(n) are input to the coefficient update unit 13. Step S120 corresponds to the second line of the algorithm.

[0043] Step S130: The coefficient update unit 13 calculates an update amount Δw of the coefficients of the adaptive filter 11. Step S130 corresponds to the third line of the algorithm, and the update amount Δw is calculated by the following formula.

[0044]

number

[0045] In the above formula, "β" is a constant to prevent the denominator from becoming 0 or too small. In this way, the update amount Δw of the coefficient of the adaptive filter 11 is calculated based on the error signal e(n) and the reference signal u(n) of the adaptive filter 11. Note that β does not have to be added.

[0046] Step S140: The coefficient update unit 13 performs a threshold determination. Step S140 corresponds to the fourth line of the algorithm, and the threshold determination is performed using the following determination formula.

[0047]

number

[0048] In the above formula, "w k "(n)" is the coefficient of the kth tap in the current adaptive filter 11, and "ψ" is a threshold value determined based on the results of verification on an actual device or the results of a simulation. Using the above judgment formula, the coefficient update unit 13 checks whether the sum of the differences between the coefficients of the current adaptive filter 11 and the coefficients of the initial transfer function for each tap, i.e., the quotient obtained by dividing the difference between the coefficients of the adaptive filter 11 and the coefficients of the initial transfer function by the sum of the coefficients of the initial transfer function ξ, is smaller than the threshold value ψ. In other words, the coefficient update unit 13 checks whether the coefficients of the current adaptive filter 11 are close to the coefficients of the initial transfer function.

[0049] If the check shows that the judgment formula holds, i.e., the coefficients of the current adaptive filter 11 are close to the coefficients of the initial transfer function (step S140: Yes), the coefficient update unit 13 executes step S150. On the other hand, if the judgment formula does not hold, i.e., the coefficients of the current adaptive filter 11 are not close to the coefficients of the initial transfer function (step S140: No), the coefficient update unit 13 executes step S160.

[0050] Step S150: The coefficient update unit 13 updates the coefficients of the adaptive filter 11 based on the coefficients of the initial transfer function. Step S150 corresponds to the fifth line of the algorithm, and the coefficients are updated by the following formula.

[0051]

number

[0052] In the above formula, "μ" is a constant indicating the step size in updating the coefficients. Also, "α" is a constant that determines the degree of influence of the difference between the current adaptive filter coefficients and the initial transfer function coefficients, and is determined in advance to a value smaller than 1 based on the results of verification on an actual device or the results of simulation. The coefficients of the adaptive filter 11 updated in this way will be used for processing at the next time point n+1.

[0053] Step S160: The coefficient update unit 13 updates the coefficients of the adaptive filter 11 without taking into account the coefficients of the initial transfer function. Step S160 corresponds to the seventh line of the algorithm, and specifically, the coefficients are updated by the following formula.

[0054]

number

[0055] After completing the above steps, the coefficient update unit 13 returns to step S120 and executes the steps from this step onwards for the input signal at the next time. The coefficient update unit 13 repeatedly executes the steps from step S120 onwards until the power of the hearing aid 100 is turned off.

[0056] [Measurement of the initial transfer function coefficients] Fig. 3 is a block diagram showing a configuration related to measurement of the coefficient W0 of the initial transfer function. In Fig. 3, the same components as those in the block diagram of the first embodiment shown in Fig. 1 are denoted by the same reference numerals. Note that the description of the common components will be omitted as appropriate.

[0057] To measure the coefficients of the initial transfer function, a hearing aid employing the NLMS algorithm as an adaptive algorithm is used, similar to the hearing aid 100 (feedback canceller 10) of the first embodiment. This hearing aid is used to measure the impulse response of the acoustic feedback unique to each hearing aid wearer. That is, the impulse response of the acoustic feedback unique to each hearing aid wearer in the time domain corresponds to the initial transfer function in the frequency domain, and the two are essentially synonymous. In summary, the impulse response expressed in the frequency domain corresponds to the feedback transfer function, the impulse response expressed in the time domain (adaptive filter coefficients) corresponds to the coefficients of the initial transfer function, and the adaptive filter coefficients correspond to the impulse response. The measurement flow will be described below.

[0058] The probe noise used for measurement is generated by the probe noise generator 90 using an M-sequence signal. The generated probe noise p(n) passes through transfer characteristics specific to the amplifier 70 and the earphone 80 and is output as sound x(n) from the earphone 80, and a part of it passes through a feedback transfer function F(z) and is fed back to the microphone 50 as feedback sound f(n).

[0059] A composite sound consisting of an external sound s(n) and a feedback sound f(n) is input to the microphone 50. At this time, it is assumed that the sound pressure of f(n) is sufficiently high and the influence of the external sound s(n) can be ignored to a certain extent. When the input composite sound passes through the transfer characteristics specific to the microphone 50 and is output as a signal d(n), the output signal y(n) of the adaptive filter 11 is subtracted from the output signal d(n) of the microphone by the subtraction unit 12, and the error signal e(n) obtained by this is input to the coefficient update unit 13.

[0060] In addition to the error signal e(n), the coefficient update unit 13 also receives a signal u(n) that has been delayed by a delay processing unit 15. The coefficient update unit 13 repeatedly updates the coefficients of the adaptive filter 11 using the NLMS algorithm so as to minimize the error signal, and when the adaptive filter 11 has sufficiently converged, the final coefficient w(n) of the adaptive filter 11 is stored as the coefficient W0 of the initial transfer function. This completes one measurement.

[0061] When measuring the coefficient W0 of the initial transfer function, various controls are performed by a measurement control unit (not shown). First, to prevent the ambient noise from significantly affecting the measurement results, the measurement control unit estimates the level of the ambient noise and its influence prior to measurement, and adjusts the sound pressure (signal strength) of the probe noise used in the measurement based on the result.

[0062] In order to prevent accidental events from significantly affecting the measurement results, the measurement control unit performs the measurement multiple times according to the above flow and determines whether to end the measurement depending on the status of each measurement result. Specifically, for example, if the difference between two consecutive measurement results or the majority of the measurement results is within a predetermined range (tolerable), the measurement is ended, whereas if it exceeds the predetermined range (unacceptable), the measurement is allowed to continue.

[0063] However, the longer the measurement takes, the greater the time and mental burden on the hearing aid wearer. Therefore, when the number of measurements exceeds a predetermined maximum number, the measurement control unit alerts the person taking the measurement (such as a hearing aid fitter) and ends the measurement.

[0064] It is also anticipated that there may be cases where the measurement must be interrupted (temporarily terminated) if the difference between the measurement results does not fall within a predetermined range due to external factors, etc. For this reason, even if it is decided to continue the measurement, the measurement can be interrupted by the measurement person (e.g., hearing aid fitter)'s operation.

[0065] The saved initial transfer function coefficient W0 is then stored in the initial value memory unit 14 of the hearing aid 100 (feedback canceller 10) after the final measurement is completed, and is used by the coefficient update unit 13 to update the coefficient of the adaptive filter 11.

[0066] Comparative Example FIG. 4 is a block diagram showing, as a comparative example, the configuration of a hearing aid 100' that includes a conventional general feedback canceller 10' that employs the NLMS algorithm.

[0067] The hearing aid 100' (feedback canceller 10') of the comparative example differs from the hearing aid 100 (feedback canceller 10) of the first embodiment in that the coefficient W0 of the initial transfer function is not used to update the coefficient of the adaptive filter 11' by the coefficient update unit 13' and in that it does not have an initial value memory unit.

[0068] 5 is a diagram showing an example of a procedure of coefficient update processing in a comparative example, in which (A) is a flowchart showing an example of the procedure of coefficient update processing executed by coefficient update unit 13', and (B) is a specific algorithm of the example of the procedure.

[0069] In the flowchart, the same steps as in the coefficient update process of the first embodiment shown in Fig. 2 are denoted by the same reference numerals. As shown in Fig. 5(A), the coefficient update process of the comparative example is composed of steps S120, S130, and S160. Step S120 corresponds to the first line of the algorithm, step S130 corresponds to the second line of the algorithm, and step S160 corresponds to the third line of the algorithm.

[0070] Compared with the coefficient update process of the first embodiment shown in Fig. 2, the coefficient update process of the comparative example does not include a step of performing a threshold determination (step S140 in Fig. 2(A)). Therefore, the coefficient update unit 13' updates the coefficients of the adaptive filter 11' in the same step (step S160) every time without taking into account the coefficients of the initial transfer function.

[0071] Thus, in the comparative example, the coefficients of the initial transfer function are not used to update the coefficients of the adaptive filter 11'. Therefore, in the hearing aid 100', when the level of the input signal is low in a quiet environment or when the closed loop gain is low, the update of the coefficients of the adaptive filter 11' is delayed, and the state in which the estimation error is large continues. In such a state, when a sound such as a pulse sound having power in the entire frequency band is suddenly input, the above-mentioned oscillation condition of howling is satisfied, and a temporary howling sound is generated from the hearing aid 100'. In addition, when a periodic signal is input in a state in which the estimation error is large, entrainment may occur.

[0072] In contrast, in the first embodiment described above, the coefficients of the initial transfer function measured at the time of initial fitting of the hearing aid are used to update the coefficients of the adaptive filter 11, and when the difference between the current coefficients of the adaptive filter 11 and the coefficients of the initial transfer function is smaller than a predetermined value (when the current coefficients are close to the coefficients of the initial transfer function), the coefficients are updated taking into account the initial transfer function, whereas when the difference is equal to or greater than the predetermined value (when the current coefficients are not close to the coefficients of the initial transfer function), the coefficients are updated without taking into account the initial transfer function. By updating the coefficients of the adaptive filter 11 in this manner, even when the input signal level is low in a quiet environment or the closed loop gain is low, the update of the adaptive filter 11 is not delayed, and as a result, the estimation error is reduced, making it possible to improve the estimation accuracy and reliably suppress howling.

[0073] Second Embodiment 6 is a block diagram showing the configuration of a hearing aid 200 including a feedback canceller 20 of the second embodiment. Description of the configuration common to the first embodiment will be omitted as appropriate.

[0074] 6, the feedback canceller 20 is composed of, for example, an adaptive filter 21, a subtraction section 22, a coefficient update section 23, an initial value storage section 24, a delay processing section 25, and band-pass filters 26, 27, 28. In this way, the hearing aid 200 (feedback canceller 20) of the second embodiment differs from the hearing aid 100 (feedback canceller 10) of the first embodiment in that it further includes three band-pass filters (hereinafter sometimes abbreviated as "BPF"). Note that in the hearing aid 200, the hearing aid processing section 65 also has the function of an amplifier section, and the hearing aid processing section 65 amplifies the gain in conjunction with various hearing aid processes, so no separate amplifier section is provided.

[0075] Of the three band-pass filters 26, 27, and 28, the first BPF 26 is provided downstream of the delay processing unit 25 and performs band limitation on the cancellation signal by the adaptive filter 21. h1 (n) is input to the adaptive filter 21 and the second BPF 27.

[0076] The second BPF 27 is provided between the first BPF 26 and the coefficient update unit 23, and the third BPF 28 is provided between the subtraction unit 22 and the coefficient update unit 23. These band-pass filters 27 and 28 limit the frequency band in order to stabilize the estimate by the adaptive filter 21. The signal u that has passed through the second BPF 27 is h (n), and the signal e passed through the third BPF 28 h (n) are all input to the coefficient update unit 23.

[0077] 7 is a diagram showing an example of the procedure of coefficient update processing in the second embodiment, in which (A) is a flowchart showing an example of the procedure of coefficient update processing executed by the coefficient update unit 23, and (B) is a specific algorithm of the example of the procedure.

[0078] In the flowchart, the same steps as those in the coefficient update process of the first embodiment shown in Fig. 2 are denoted by the same reference numerals. As shown in Fig. 7(A), the coefficient update process of the second embodiment differs from that of the first embodiment only in the third step (step S130 in the first embodiment is replaced by step S230 in the second embodiment). Step S230 corresponds to the third line of the algorithm. In step S230, the coefficient update unit 23 calculates the update amount Δw of the coefficient of the adaptive filter 21 by the following formula.

[0079]

number

[0080] In this way, in the second embodiment, the signal u h (n), and the signal e passed through the third BPF 28 h (n), that is, the reference signal u of the adaptive filter 21 h1 (n) is band-limited signal u h (n), and the error signal e(n) band-limited signal e h Based on (n), the update amount Δw of the coefficient of the adaptive filter 21 is calculated, and then threshold determination is performed in the same manner as in the first embodiment. By combining an adaptive filter using the NLMS algorithm with band limitation by a bandpass filter, howling can be suppressed more stably than in the first embodiment. Furthermore, by updating the coefficient of the adaptive filter 21 in the above-mentioned manner, the estimation error is smaller than in the first embodiment, so that the estimation accuracy can be further improved and howling can be suppressed more reliably.

[0081] Third Embodiment 8 is a block diagram showing the configuration of a hearing aid 300 including a feedback canceller 30 of the third embodiment. Description of the configuration common to the first embodiment will be omitted as appropriate.

[0082] As shown in Fig. 8, the feedback canceller 30 is composed of, for example, an adaptive filter 31, a subtraction section 32, a coefficient update section 33, an initial value storage section 34, a delay processing section 35, and adaptive lattice filters 36 and 37. In this way, the hearing aid 300 (feedback canceller 30) of the third embodiment differs from the hearing aid 100 (feedback canceller 10) of the first embodiment in that it further comprises two adaptive lattice filters (hereinafter sometimes abbreviated as "ALF"). Note that in the hearing aid 300, the hearing aid processor 65 also has the function of an amplifier section, and the hearing aid processor 65 amplifies the gain in addition to various hearing aid processes, so no separate amplifier section is provided.

[0083] Of the two adaptive lattice filters 36 and 37, one ALF 36 is provided between the delay processing unit 35 and the coefficient update unit 33, and whitens the signal u(n) after the delay processing. The other ALF 37 is provided between the subtraction unit 32 and the coefficient update unit 33, and whitens the error signal e(n). The signal u(n) whitened by the ALF 36 p (n) and the signal whitened with ALF37 e p (n) are all input to the coefficient update unit 33.

[0084] 9 is a diagram showing an example of the procedure of coefficient update processing in the third embodiment, in which (A) is a flowchart showing an example of the procedure of coefficient update processing executed by the coefficient update unit 33, and (B) is a specific algorithm of the example of the procedure.

[0085] In the flowchart, the same steps as those in the coefficient update process of the first embodiment shown in Fig. 2 are denoted by the same reference numerals. As shown in Fig. 9(A), the coefficient update process of the third embodiment differs from that of the first embodiment only in the third step S330 (step S130 in the first embodiment is replaced by step S330 in the second embodiment). Step S330 corresponds to the third line of the algorithm. In step S330, the coefficient update unit 33 calculates the update amount Δw of the coefficient of the adaptive filter 31 by the following formula.

[0086]

number

[0087] In this way, in the third embodiment, the signal u p (n) and the signal e passed through ALF37 p (n), that is, the signal u(n) obtained by performing whitening processing on the reference signal u(n) of the adaptive filter 31. p (n), and the error signal e(n) subjected to whitening processing p Based on (n), the update amount Δw of the coefficient of the adaptive filter 31 is calculated, and then threshold determination is performed in the same manner as in the first embodiment. By combining the adaptive filter using the NLMS algorithm with whitening by an adaptive lattice filter, it is possible to suppress howling more stably than in the first embodiment. Furthermore, by updating the coefficient of the adaptive filter 31 in the above-mentioned manner, the estimation error is smaller than in the first embodiment, and therefore the estimation accuracy can be further improved.

[0088] [Comparison of estimation errors by simulation] Fig. 10 is a diagram showing a comparison of the simulation results of the feedback canceller of the embodiment and the comparative example. In Fig. 10, the vertical axis is the estimation error, and the estimation error of the adaptive filter decreases toward the bottom. The thick line represents the simulation result when the feedback canceller 30 of the third embodiment is used, and the thin line represents the simulation result when the feedback canceller 10' of the comparative example is used.

[0089] As shown in FIG. 10, the feedback canceller 30 of the third embodiment had a smaller estimation error than the feedback canceller 10' of the comparative example for most of the time after the start of the simulation.

[0090] While the simulation was running, continuous environmental sounds with little fluctuation, such as a relatively quiet air conditioner operating sound, were input to the system. The estimation error was large at the beginning of the simulation, but after a short time the adaptive algorithm worked to reduce the estimation error, and after about 2.5 seconds from the start it had basically leveled off. This shows that as long as the input sound does not change significantly, feedback can be stably suppressed.

[0091] Also, at the beginning of the simulation, there is almost no difference in the estimation error between the third embodiment and the comparative example, but the difference starts to widen from about 0.5 seconds after the start, and thereafter the third embodiment continues to maintain a smaller estimation error than the comparative example. This shows that the third embodiment has higher estimation accuracy than the comparative example, and can remove the feedback component more accurately.

[0092] In addition, a colored signal, specifically the sound of a so-called bell (a Buddhist instrument struck on a bell stand with a stick), was input to the system for approximately two seconds starting six seconds after the start of the simulation. Therefore, the estimation error rose sharply temporarily at this timing, but then dropped again in a short time. Also, even during this time period, the third embodiment was able to maintain a smaller estimation error than the comparative example. From this, it can be seen that even when the input sound changes significantly, the third embodiment has a higher estimation accuracy than the comparative example, and is capable of reducing the probability of entrainment occurring due to periodic signals.

[0093] In this way, the feedback canceller 30 of the third embodiment has a lower estimation error than the comparative example, even when the input sound does not change significantly, or when the input sound changes significantly, and has higher estimation accuracy and can more stably remove the feedback component. Therefore, the superiority of the embodiment is clear.

[0094] [Advantages of the present invention] As described above, according to the above-described embodiment, the following effects can be obtained.

[0095] (1) The coefficients of the adaptive filter are updated using the impulse response (coefficients of the initial transfer function) when the hearing aid is first worn, so that the update of the adaptive filter is not delayed even in quiet environments or when the loop gain is low, resulting in a smaller estimation error (improved estimation accuracy) and reliable suppression of howling. It is also expected that the occurrence of entrainment due to estimation error can be suppressed.

[0096] (2) By measuring the coefficients of the initial transfer function at the time of initial fitting of the hearing aid, it is possible to accommodate individual differences in the impulse response coefficients between hearing aid wearers (such as the shape and softness of the pinna).

[0097] The present invention is not limited to the above-described embodiment, and can be implemented in various modified forms.

[0098] In each of the above-described embodiments, the feedback canceller is installed in a hearing aid, but this is not limited to this and it can be installed in various audio devices equipped with a sound input means (microphone) and an output means (speaker or earphones), such as a loudspeaker or karaoke machine.

[0099] In each of the above-described embodiments, the constant γ used in calculating the sum ξ of the coefficients of the initial transfer function in the first step of the coefficient update process (step S110 in Figs. 2, 7, and 9) is determined uniformly based on the results of verification using an actual device or the results of a simulation, but instead, it may be set to a different value for each hearing aid wearer. With this configuration, the true decision rate in the threshold decision can be optimally adjusted according to the level of the impulse response for each hearing aid wearer, making it possible to further improve the estimation accuracy and more reliably suppress howling.

[0100] In each of the above-described embodiments, the step size used when updating the coefficients of the adaptive filter is a constant. However, instead of this, the step size may be a variable value.

[0101] In the above-described second embodiment, a band-pass filter is used to limit the frequency band of the signal, but instead of this, for example, a high-pass filter may be used.

[0102] In the above-described third embodiment, an adaptive lattice filter is used to whiten the signal, but the present invention is not limited to this and other techniques may be used.

[0103] Furthermore, the configurations and numerical values ​​given in the course of explaining the feedback cancellers 10, 20, 30 and the hearing aids 100, 200, 300 are merely examples, and it goes without saying that suitable modifications can be made when implementing the present invention. [Explanation of symbols]

[0104] 10 Feedback Canceller 11 Adaptive Filters 12 Subtraction section 13 Coefficient update section 14 Initial value storage section 15 Delay Processing Section 50 Microphone 60 Hearing aid processor 70 Amplification section 80 Earphones 100 Hearing aids

Claims

1. an adaptive filter that estimates an impulse response from a sound output means to a sound input means and outputs a signal corresponding to an estimate of a feedback component based on a reference signal input to the adaptive filter; a subtraction unit which outputs an error signal obtained by subtracting the output signal of the adaptive filter from the output signal of the sound input means; a storage unit that stores coefficients of an initial transfer function based on a feedback impulse response specific to a usage environment, the feedback impulse response being measured in advance using the sound output means and the sound input means; a coefficient updating unit that calculates an update amount of the coefficient of the adaptive filter based on the reference signal and the error signal, and updates the coefficient of the adaptive filter taking into account the coefficient of the initial transfer function in the update amount when a difference between a current coefficient of the adaptive filter and a coefficient of the initial transfer function is smaller than a predetermined value, and updates the coefficient of the adaptive filter without taking into account the coefficient of the initial transfer function when the difference is equal to or larger than the predetermined value; A feedback canceller comprising:

2. 2. The feedback canceller of claim 1, a band limiting unit that limits a frequency band of at least one of the reference signal and the error signal; Further equipped with The coefficient update unit A feedback canceller comprising: a filter for filtering a signal from a band limiter; a filter for filtering a signal from the filter;

3. 2. The feedback canceller of claim 1, a whitening unit for whitening at least one of the reference signal and the error signal; Further equipped with The coefficient update unit A feedback canceller comprising: a whitening unit that calculates an amount of update for a coefficient of the adaptive filter based on an output signal of the whitening unit.

4. 4. The feedback canceller according to claim 1, The coefficient update unit a feedback canceller for updating the current coefficients by adding a value obtained by multiplying a difference between a current coefficient of the adaptive filter and a coefficient of the initial transfer function by a predetermined constant and subtracting the result from the update amount when the difference is smaller than a predetermined value, and for updating the current coefficients by adding the update amount when the difference is equal to or greater than the predetermined value.

5. at least, A feedback canceller according to any one of claims 1 to 4, a sound input unit that converts an input sound into an electrical signal and outputs the electrical signal to the subtraction unit; a sound output means for converting an electrical signal based on the error signal into sound and outputting the sound; Equipped with audio equipment.

6. A measuring step of measuring a coefficient of an initial transfer function based on an impulse response of feedback specific to a usage environment with respect to an impulse response from a sound output means to a sound input means; an estimation step of estimating the impulse response and outputting a signal corresponding to an estimated value of a feedback component based on an input reference signal; a subtraction step of outputting an error signal obtained by subtracting the signal output in the estimation step from the output signal of the sound input means; a coefficient updating step of calculating an update amount of an estimation coefficient used for estimation in the estimation step based on the reference signal and the error signal, and updating the estimation coefficient by taking the coefficient of the initial transfer function into consideration in the update amount when a difference between the current estimation coefficient and the coefficient of the initial transfer function is smaller than a predetermined value, and updating the estimation coefficient without taking the coefficient of the initial transfer function into consideration when the difference is equal to or larger than the predetermined value; How to cancel feedback, including:

Images