CN111916099B - Adaptive echo cancellation device and method for variable-step hearing aid - Google Patents
Adaptive echo cancellation device and method for variable-step hearing aid Download PDFInfo
- Publication number
- CN111916099B CN111916099B CN202011090523.1A CN202011090523A CN111916099B CN 111916099 B CN111916099 B CN 111916099B CN 202011090523 A CN202011090523 A CN 202011090523A CN 111916099 B CN111916099 B CN 111916099B
- Authority
- CN
- China
- Prior art keywords
- energy
- frequency
- parameter
- adaptive
- time
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L21/00—Processing of the speech or voice signal to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
- G10L21/02—Speech enhancement, e.g. noise reduction or echo cancellation
- G10L21/0208—Noise filtering
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L21/00—Processing of the speech or voice signal to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
- G10L21/02—Speech enhancement, e.g. noise reduction or echo cancellation
- G10L21/0208—Noise filtering
- G10L2021/02082—Noise filtering the noise being echo, reverberation of the speech
Abstract
The invention discloses a self-adaptive echo cancellation device and a self-adaptive echo cancellation method for a variable-step-size hearing aid, wherein the self-adaptive echo cancellation device comprises a single-frequency tone detector, a step-size controller and a self-adaptive filter: the single-frequency tone detector is used for carrying out spectrum energy analysis on the error signal sample, calculating the current state parameter of the hearing aid system and transmitting the parameter to the step length controller; the step length controller is used for judging the state of the system according to the system state parameter and the normalization correction coefficient obtained from the single-frequency tone detector, calculating a time-varying step length parameter of the self-adaptive filter and transmitting the time-varying step length parameter to the self-adaptive filter; the self-adaptive filter is used for filtering the cached far-end signal sample of the loudspeaker, calculating and estimating an echo signal and then outputting the echo signal, and iteratively updating the self-adaptive filter according to the time-varying step length parameter calculated by the step length controller. The method solves the problems that the self-adaptive filter in the prior art is low in convergence speed and high in algorithm complexity and is difficult to realize, and is suitable for a digital hearing aid system.
Description
Technical Field
The invention relates to a step-size-variable hearing aid self-adaptive echo cancellation device and a step-size-variable hearing aid self-adaptive echo cancellation method.
Background
With the development of industrialization and the increasing aging of recent years, the number of people suffering from hearing loss diseases is continuously increasing, and the selection of hearing aids to improve hearing loss is the most common and quick auxiliary means. However, the hearing aid has a very easy structure and function to amplify echo and generate howling, so the echo cancellation algorithm is one of the key algorithms of the digital hearing aid, which estimates the echo signal by estimating the characteristic parameters of the echo path, using the far-end signal of the loudspeaker as a reference, and then estimates the echo signal from the received near-end microphone signal. Since the echo path is usually unknown and time-varying, especially since slight wear adjustments during use of the hearing aid change the echo path, the adaptive filter becomes a key component of the hearing aid echo cancellation module by virtue of its ability to track the time-varying system.
The patent with publication number CN101179294B discloses an adaptive echo canceller and an echo cancellation method thereof, and as described in the background of paragraph [0011] of the present application, "in an embedded system, an adaptive algorithm commonly used in an echo canceller is an algorithm group based on the steepest descent method. A representation of such an adaptive algorithm is: LMS (least mean square error) algorithm, whose minimization criterion is the root mean square error. The self-adaptive algorithm has the advantages of small calculation amount, strong robustness and easy realization, and is widely adopted in practice. The disadvantages are: the convergence speed is slow and the convergence performance is sensitive to energy variations of the input signal. The NLMS (energy normalized minimum mean square error) algorithm is an improved algorithm of the LMS algorithm and overcomes the defect that the LMS algorithm is sensitive to the energy of an input signal. The NLMS algorithm and its various modifications are the adaptive filtering algorithm in echo cancellers that are mainly used at present.
Taking an adaptive algorithm LMS as an example, in the adaptive process of the filter, in practical situations, especially in the field of hearing aids, noise or voice signals inevitably exist in the near-end signal except for the desired signal, and the noise equivalent to the noise added with a large signal greatly affects the convergence rate of the adaptive process and may cause divergence in severe cases, and this condition is called double-ended sounding (DT) in the hearing aid echo cancellation system. The misadjustment of the self-adaptation process of the filter can be slowed down to a certain extent by directly reducing the convergence step size parameter, however, the convergence speed of the algorithm and the tracking speed of the time-varying system can be seriously reduced, and the calculation difficulty in the fixed-point system can be caused by the too small step size parameter.
Some echo cancellation algorithms include a double talk detection module (DTD): the double talk detection is to determine the talk state of the current system (near-end talk, far-end talk, double-end talk), when the system is determined to be double-end talk, the adaptive filter does not update the coefficient, and the general DTD method has a disadvantage that the method cannot adapt to the characteristic that the echo path in the echo cancellation environment is changed continuously. Therefore, many scholars have actively studied the existing problems in the adaptive filter echo cancellation method and have proposed many improvements. The method mainly comprises the steps of giving a variable step function or analyzing an input end signal and the like to control the updating of the coefficient of the adaptive filter so as to improve the stable initial convergence speed and the stable error of an echo path. However, these methods generally have the problems of high requirement on computational accuracy and high computational complexity, and are difficult to implement in a hearing aid echo cancellation system, and in the prior art, there is no method suitable for a hearing aid and considering power consumption and computational complexity, which can solve the above problems well.
Disclosure of Invention
Aiming at the problems, the invention provides a variable step length hearing aid self-adaptive echo cancellation device and an echo cancellation method, which solve the problems that the self-adaptive filter in the prior art is low in convergence speed and too high in algorithm complexity and difficult to realize; furthermore, the problems that the steady-state error of the self-adaptive filter is large, double-end sounding and parameter misadjustment can be caused by related data in the prior art are solved.
In order to achieve the technical purpose and achieve the technical effect, the invention is realized by the following technical scheme:
a step size-variable hearing aid adaptive echo cancellation device comprises a single-frequency tone detector, a step size controller and an adaptive filter:
the single-frequency tone detector is used for carrying out spectrum energy analysis on an error signal e (n) sample, calculating a current hearing aid system state parameter and transmitting the parameter to the step length controller;
the step length controller is used for judging the state of the system according to the system state parameter obtained from the single-frequency tone detector and the normalization correction coefficient f (m), calculating a time-varying step length parameter mu (m) of the self-adaptive filter and transmitting the time-varying step length parameter mu (m) to the self-adaptive filter;
the adaptive filter is used for filtering the buffered loudspeaker far-end signal u (n) samples and calculating an estimated echo signal
Then outputting, and carrying out iterative updating on the self-adaptive filter according to the time-varying step length parameter calculated by the step length controller;
wherein the error signal e (n) is the near-end speech signal v (n) minus the estimated echo signal
Preferably, the single-frequency tone detector includes a framing FFT module and a spectral energy analysis module:
the frame-dividing FFT module is used for receiving the cached error signal e (n), carrying out fast Fourier transform on the error signal e (n) according to a set frequency, calculating the power spectrum of the signal and transmitting the power spectrum to the spectrum energy analysis module;
the spectrum energy analysis module analyzes the maximum single-frequency energy P of the obtained signal spectrummax(m) the ratio (m) of the maximum single-frequency energy to the total energy of the power spectrum, and transmitting the obtained system state parameters to the step length controller.
Preferably, the step size controller comprises a state selection module, and the state selection module selects the state of the system according to the system state parameters obtained by the single-frequency tone detector, controls and adjusts the time-varying step size parameter μ (m), and transmits the time-varying step size parameter μ (m) to the adaptive filter.
Preferably, the step size controller obtains the maximum single-frequency energy P according to the normalized correction coefficient f (m) and the single-frequency tone detectormax(m) judging a threshold value according to the ratio (m) of the maximum single-frequency energy to the total energy of the power spectrum, and controlling and adjusting a time-varying step length parameter mu (m) according to a judgment result, wherein the judgment result specifically comprises the following steps:
1) if the ratio (m) of the maximum single-frequency energy to the total energy of the power spectrum is smaller than a set threshold Th2, the system is judged to be in steady state convergence, and a preset steady state step length parameter mu is adoptedaAs a time-varying step size parameter;
2) if the ratio (m) of the maximum single-frequency energy to the total energy of the power spectrum is greater than a set threshold Th2, and the maximum single-frequency energy Pmax(m) is less than a set threshold Th3, the system is determined to be in a relevant interference state, and the time-varying step parameter is corrected by using a normalized correction coefficient f (m): μ (m) ═ f (m) · μa；
3) If the ratio (m) of the maximum single-frequency energy to the total energy of the power spectrum is greater than a set threshold Th2, and the maximum single-frequency energy Pmax(m) is greater than a set threshold Th3, and the normalization correction coefficient f (m) is less than a set threshold Th4, the system is determined to be in a howling state, at the moment, howling is preferentially processed, and a fixed step size parameter mu is adoptedbFilter iteration is accelerated: μ (m) ═ μbWherein, mub＞μa；
4) If the ratio (m) of the maximum single-frequency energy to the total energy of the power spectrum is greater than a set threshold Th2, and the maximum single-frequency energy Pmax(m) is greater than a set threshold Th3, and the normalized correction factor f (m) is greater than a set threshold Th4, then the system is determined to be in a double-ended sounding state, at which time the time-varying step size parameter is set to zero to stop the adaptive iteration of the filter: μ (m) ═ 0.
Preferably, the adaptive filter includes a normalized error signal processing module, a coefficient updating module, and a FIR filter module:
the normalization error signal processing module normalizes the square of the Euclidean norm of the error signal e (n) relative to the loudspeaker far-end signal u (n) to obtain a normalization error signal
Wherein, U (n) ═ { U (n), U (n +1),.. U (n + D-1) } is the far-end signal vector corresponding to the time delay, D is the adaptive filter order and is a constant;
normalizing error signal e according to set frequency by convex combination first-order recursion processnor m(n) performing snapshot smoothing to obtain a normalized error mean value T (m);
the coefficient updating module is used for updating the coefficient according to the normalized error signal enor m(n), the normalized error mean value T (m) and the time-varying step size parameter mu (m) obtained by the step size controller, and the adaptive filter coefficient is subjected to iterative updating;
the FIR filter module carries out filtering processing on the far-end signal u (n) to obtain an estimated echo signal
And then outputting.
A variable step size hearing aid adaptive echo cancellation device according to any one of the preceding claims, adapted for use in a digital hearing aid system.
Correspondingly, the adaptive echo cancellation method for the variable-step hearing aid comprises the following steps:
A. according to the cached far-end signal sample, the estimation value is obtained through filtering of a self-adaptive FIR filterEcho signal of meter
Wherein W (n) { W ═ Wn(1)，wn(2)，...，wn(D)}，wn(x) FIR filter coefficients, WH(n) is the conjugate transpose of W (n); and subtracting the estimated echo signal from the sampled near-end speech signal v (n)
Obtaining an error signal
U (n) ═ U (n), U (n +1),.. U (n + D-1) } is the far-end signal vector corresponding to the time delay, D is the adaptive filter order;
B. performing time-frequency transformation on the error signal, calculating the power spectrum of the error signal and analyzing the energy distribution characteristic to obtain system state parameters;
C. judging the state of the system according to the system state parameters, and controlling the self-adaptive time-varying step length parameter mu (m);
D. normalizing the squared Euclidean norm of the error signal e (n) relative to the far-end signal u (n) of the loudspeaker to obtain a normalized error signal enor m(n), smoothing the maximum value of each frame of the normalized error signal by using a convex combination first-order recursion process, and taking the maximum value as an upper limit threshold of the normalized adaptive error;
E. using the corrected time-varying step size parameter mu (m), normalizing the error signal enor m(n) iteratively updating FIR filter coefficients: w (n) ═ W (n-1) + μ (m) · enorm(n)·U(n)。
Preferably, the step B specifically includes:
b1, continuously filling error signals into a buffer area of the single-frequency tone detector, performing one-time fast Fourier transform operation after the buffer area is full of data, resetting the buffer to obtain the frequency spectrum of the signals, and further calculating the power spectrum PSD of the signalsm(k)：
PSDm(k)＝|fft(E(m))|2Wherein, m is the frame number, k is the frequency point, E (m) ═ { E (N), E (N +1),.., E (N + N-1) }; FFT (—) is the FFT operation; n is the number of fast Fourier transform points;
b2, searching the maximum energy frequency point k according to the power spectrum of the signalmaxCalculating the energy sum P of the maximum energy frequency point and the adjacent frequency pointsmax(m)：
Pmax(m)＝PSDm(kmax-1)+PSDm(kmax)+PSDm(kmax+1)；
Calculating the ratio (m) of the maximum single-frequency energy to the total energy of the power spectrum:
wherein N is the number of fast Fourier transform points;
the state parameter Pmax(m) and (m) are transmitted to the step size controller.
Preferably, the step C specifically includes:
c1, according to the ratio (m) of the maximum single-frequency energy obtained by the single-frequency tone detector to the total energy of the power spectrum, smoothing and normalizing the correction coefficient f (m):
f(m+1)＝λ·f(m)+(1-λ)·I(m)；
wherein, λ is a forgetting factor and is a positive number smaller than 1;
c2 maximum single frequency energy Pmax(m), the ratio (m) of the maximum single-frequency energy to the total energy of the power spectrum and a normalization correction coefficient f (m) are subjected to threshold judgment, and according to the judgment results of the three state parameters, a time-varying step length parameter is adjusted, specifically:
1) if the ratio (m) of the maximum single-frequency energy to the total energy of the power spectrum is less thanDetermining a threshold Th2, determining the system to be in steady state convergence, and adopting a preset steady state step size parameter muaAs a time-varying step size parameter;
2) if the ratio (m) of the maximum single-frequency energy to the total energy of the power spectrum is greater than a set threshold Th2, and the maximum single-frequency energy Pmax(m) is less than a set threshold Th3, the system is determined to be in a relevant interference state, and the time-varying step parameter is corrected by using a normalized correction coefficient f (m): μ (m) ═ f (m) · μa；
3) If the ratio (m) of the maximum single-frequency energy to the total energy of the power spectrum is greater than a set threshold Th2, and the maximum single-frequency energy Pmax(m) is greater than a set threshold Th3, and the normalization correction coefficient f (m) is less than a set threshold Th4, the system is determined to be in a howling state, at the moment, howling is preferentially processed, and a fixed step size parameter mu is adoptedbFilter iteration is accelerated: μ (m) ═ μbWherein, mub＞μa；
4) If the ratio (m) of the maximum single-frequency energy to the total energy of the power spectrum is greater than a set threshold Th2, and the maximum single-frequency energy Pmax(m) is greater than a set threshold Th3, and the normalized correction factor f (m) is greater than a set threshold Th4, then the system is determined to be in a double-ended sounding state, at which time the time-varying step size parameter is set to zero to stop the adaptive iteration of the filter: μ (m) ═ 0.
Preferably, the step D specifically includes:
d1, squaring U (n) of the Euclidean norm of the current error signal e (n) relative to the speaker far-end signal U (n)2Normalization is carried out to obtain a normalized error signal enor m(n)：
Wherein, is a constant;
d2, obtaining absolute value | e of normalized error signal of each framenor mMaximum value e of (n) |max(m) smoothing said data with a convex combination first order recursive process to obtain a normalized error mean T (m) of the normalized error signal:
wherein, alpha and beta are forgetting factors;
d3, using the normalized error mean T (m) as the normalized error signal enor m(n) upper threshold limit.
The invention has the beneficial effects that:
1) the invention takes the detection data of the single-frequency tone detector as the basis, and by means of the multiplier, FFT and other modules of the hearing aid chip, the algorithm is low in complexity and power consumption, is easy to realize in the digital hearing aid, and solves the problems that the self-adaptive filter in the prior art is low in convergence speed and high in algorithm complexity and is difficult to realize.
2) According to the method, the iteration step length is corrected according to the distribution characteristics of the frequency spectrum energy of the historical data after echo elimination to weaken the maladjustment problem caused by single-frequency noise, the working state of the adaptive filter is switched to quickly inhibit howling, good balance is achieved in the convergence speed and the steady-state error of the adaptive filter, and a time-varying system can be quickly tracked and howling inhibition can be quickly achieved.
3) The invention takes the long-term average value of the normalized error signal as the dynamic threshold value to control the iteration step length of the filter not to have sudden change, weakens the influence of double-end talkback under the conditions of higher robustness and faster convergence speed, and solves the problems of large steady-state error of the self-adaptive filter, double-end sounding, parameter imbalance generation of related data and the like in the prior art.
4) The adaptive filter has strong anti-interference capability in an external noise or double-end talkback scene, and can obviously prevent the iterative divergence of the adaptive filter. When the device is in an environment with strong noise single-frequency energy, the detuning can be effectively prevented.
Drawings
Fig. 1 is a block diagram of an adaptive echo cancellation device for a hearing aid with variable step size according to the present invention;
fig. 2 is a diagram comparing the convergence performance of the present method and the conventional method in the presence of external interference.
Detailed Description
The present invention will be better understood and implemented by those skilled in the art by the following detailed description of the technical solution of the present invention with reference to the accompanying drawings and specific examples, which are not intended to limit the present invention.
As shown in fig. 1, a step-size-variable hearing aid adaptive echo cancellation apparatus mainly includes a single-frequency tone detector, a step-size controller, and an adaptive filter, and specifically, the following components are mainly introduced:
the single-frequency tone detector is used for carrying out spectrum energy analysis on the error signal e (n) sample, calculating the current state parameter of the hearing aid system and transmitting the parameter to the step length controller.
The adaptive echo cancellation device collects an external input signal s (n) to obtain a near-end speech signal v (n), wherein an error signal e (n) is obtained by subtracting an estimated echo signal from the near-end speech signal v (n)
The step length controller is used for judging the state of the system according to the system state parameter obtained from the single-frequency tone detector and the normalization correction coefficient f (m), calculating the time-varying step length parameter mu (m) of the self-adaptive filter and transmitting the time-varying step length parameter mu (m) to the self-adaptive filter.
The adaptive filter is used for filtering the buffered loudspeaker far-end signal u (n) samples and calculating an estimated echo signal
And then outputting, and carrying out iterative updating on the self-adaptive filter according to the time-varying step length parameter calculated by the step length controller. In the filtering process, the time-varying step size parameter updated last time is adopted for filtering, and the time-varying step size parameter updated this time is used for the next filtering process, so that the operation can be reduced, the power consumption of the hearing aid is reduced, and the implementation in the digital hearing aid is easier.
Preferably, the main structure and function of each component of the adaptive echo cancellation device are described as follows:
preferably, the single-frequency tone detector includes a framing FFT module and a spectral energy analysis module, wherein:
the frame-dividing FFT module is used for receiving the cached error signal e (n), performing Fast Fourier Transform (FFT) on the error signal e (n) according to a set frequency, calculating a power spectrum of the signal and transmitting the power spectrum to the spectrum energy analysis module;
the spectrum energy analysis module analyzes the maximum single-frequency energy P of the obtained signal spectrummax(m) the ratio (m) of the maximum single-frequency energy to the total energy of the power spectrum, and transmitting the obtained system state parameters to the step length controller.
Preferably, the step size controller comprises a state selection module, and the state selection module selects the state of the system according to the system state parameters obtained by the single-frequency tone detector, controls and adjusts the time-varying step size parameter μ (m), and transmits the time-varying step size parameter μ (m) to the adaptive filter.
For example, the step size controller obtains the maximum single-frequency energy P according to the normalized correction coefficient f (m) and the single-frequency tone detectormax(m) judging a threshold value according to the ratio (m) of the maximum single-frequency energy to the total energy of the power spectrum, and controlling and adjusting a time-varying step length parameter mu (m) according to a judgment result, wherein the judgment result specifically comprises the following steps:
1) if the ratio (m) of the maximum single-frequency energy to the total energy of the power spectrum is smaller than a set threshold Th2, the system is judged to be in steady state convergence, and a preset steady state step length parameter mu is adoptedaAs the time-varying step parameter, in general, μaThe value may be 0.05;
2) if the ratio (m) of the maximum single-frequency energy to the total energy of the power spectrum is greater than a set threshold Th2, and the maximum single-frequency energy Pmax(m) is less than a set threshold Th3, the system is determined to be in a relevant interference state, and the time-varying step parameter is corrected by using a normalized correction coefficient f (m): μ (m) ═ f (m) · μa；
3) If the ratio (m) of the maximum single-frequency energy to the total energy of the power spectrum is greater than a set threshold Th2, and the maximum single-frequency energy Pmax(m) is greater than a set threshold value Th3, and the normalization correction coefficient f (m) is less than a set threshold value Th4, the system is determined to be howlingA squeal state, wherein the squeal is preferentially processed, and a larger fixed step size parameter mu is adoptedbFilter iteration is accelerated: μ (m) ═ μbWherein, mub＞μa；
4) If the ratio (m) of the maximum single-frequency energy to the total energy of the power spectrum is greater than a set threshold Th2, and the maximum single-frequency energy Pmax(m) is greater than a set threshold Th3, and the normalized correction factor f (m) is greater than a set threshold Th4, then the system is determined to be in a double-ended sounding state, at which time the time-varying step size parameter is set to zero to stop the adaptive iteration of the filter: μ (m) ═ 0.
The set threshold values Th2, Th3 and Th4 are all constants and can be obtained by regulating the specific system state.
Preferably, the adaptive filter includes a normalized error signal processing module, a coefficient updating module, and a FIR filter module, wherein:
the normalization error signal processing module normalizes the square of the Euclidean norm of the error signal e (n) relative to the loudspeaker far-end signal u (n) to obtain a normalization error signal
U (n) ═ U (n), U (n +1),.. U (n + D-1) } is the far-end signal vector corresponding to the time delay, and D is the adaptive filter order and is a small constant used for preventing divergence caused by too small denominator.
Normalizing error signal e according to set frequency by convex combination first-order recursion processnor mAnd (n) performing snapshot smoothing to obtain a normalized error mean value T (m).
And the coefficient updating module is used for iteratively updating the coefficient of the adaptive filter according to the normalized error signal, the normalized average error and the time-varying step size parameter mu (m) obtained by the step size controller.
The FIR filter module carries out filtering processing on the far-end signal u (n) to obtain an estimated echo signal
And then outputting.
Correspondingly, the hearing aid adaptive echo cancellation device according to any one of the above items is suitable for a digital hearing aid system. The digital hearing aid automatically collects the conditions of the acoustic signal type, the signal-to-noise ratio, the strength difference of the front microphone and the rear microphone and the like of the environment where the digital hearing aid is located, defines different environments, and automatically adjusts the characteristics of noise reduction, direction, compression ratio and the like so as to adapt to the continuously changing environment. The phenomenon that a user of the analog machine cannot hear the sound with small sound and is difficult to hear with loud sound is avoided. Generally, a digital hearing aid system mainly includes an echo cancellation device, a microphone (or microphone), an amplifier, a receiver (or earphone), a battery, electro-acoustic devices such as various volume or tone control knobs, and a housing, wherein the echo cancellation device can adopt any one of the above-mentioned step-size-variable hearing aid adaptive echo cancellation devices.
Correspondingly, the adaptive echo cancellation method for the variable-step hearing aid comprises the following steps:
A. according to the cached far-end signal sample, the estimated echo signal is obtained through filtering of a self-adaptive FIR filter
Wherein W (n) { W ═ Wn(1)，wn(2)，...，wn(D)}，wn(x) FIR filter coefficients, WH(n) is a conjugate transpose of W (n), U (n) ═ U (n), U (n +1),.. U (n + D-1) } is a far-end signal vector corresponding to the time delay, and D is an adaptive filter order; and subtracting the estimated echo signal from the sampled near-end speech signal v (n)
Obtaining an error signal
B. Performing time-frequency transformation on the error signal, calculating a power spectrum of the error signal, analyzing energy distribution characteristics, and obtaining system state parameters, preferably, the step B specifically includes:
b1, continuously filling error signals into a buffer area of the single-frequency tone detector, performing one-time fast Fourier transform operation after the buffer area is full of data, resetting the buffer to obtain the frequency spectrum of the signals, and further calculating the power spectrum PSD of the signalsm(k)：
PSDm(k)＝|fft(E(m))|2Wherein, m is the frame number, k is the frequency point, E (m) ═ { E (N), E (N +1),.., E (N + N-1) }; FFT (—) is the FFT operation; n is the number of fast Fourier transform points;
b2, searching the maximum energy frequency point k according to the power spectrum of the signalmaxCalculating the energy sum P of the maximum energy frequency point and the adjacent frequency pointsmax(m)：
Pmax(m)＝PSDm(kmax-1)+PSDm(kmax)+PSDm(kmax+1)；
Calculating the ratio (m) of the maximum single-frequency energy to the total energy of the power spectrum:
wherein N is the number of fast Fourier transform points;
the state parameter Pmax(m) and (m) are transmitted to the step size controller.
C. And judging the state of the system according to the system state parameters, and controlling the self-adaptive time-varying step length parameter mu (m). Preferably, the step C specifically includes:
c1, according to the ratio (m) of the maximum single-frequency energy obtained by the single-frequency tone detector to the total energy of the power spectrum, smoothing and normalizing the correction coefficient f (m):
f(m+1)＝λ·f(m)+(1-λ)·I(m)；
wherein, λ is a forgetting factor and is a positive number smaller than 1;
C2 maximum single frequency energy Pmax(m), the ratio (m) of the maximum single-frequency energy to the total energy of the power spectrum, and a normalization correction coefficient f (m) are subjected to threshold judgment, and a time-varying step length parameter is adjusted according to the judgment results of the three state parameters, wherein the preferable step C2 specifically comprises the following steps:
1) if the ratio (m) of the maximum single-frequency energy to the total energy of the power spectrum is smaller than a set threshold Th2, the system is judged to be in steady state convergence, and a preset steady state step length parameter mu is adoptedaAs a time-varying step size parameter;
2) if the ratio (m) of the maximum single-frequency energy to the total energy of the power spectrum is greater than a set threshold Th2, and the maximum single-frequency energy Pmax(m) is less than a set threshold Th3, the system is determined to be in a relevant interference state, and the time-varying step parameter is corrected by using a normalized correction coefficient f (m): μ (m) ═ f (m) · μa；
3) If the ratio (m) of the maximum single-frequency energy to the total energy of the power spectrum is greater than a set threshold Th2, and the maximum single-frequency energy Pmax(m) is greater than a set threshold Th3, and the normalization correction coefficient f (m) is less than a set threshold Th4, the system is determined to be in a howling state, at the moment, howling is preferentially processed, and a fixed step size parameter mu is adoptedbFilter iteration is accelerated: μ (m) ═ μbWherein, mub＞μa；
4) If the ratio (m) of the maximum single-frequency energy to the total energy of the power spectrum is greater than a set threshold Th2, and the maximum single-frequency energy Pmax(m) is greater than a set threshold Th3, and the normalized correction factor f (m) is greater than a set threshold Th4, then the system is determined to be in a double-ended sounding state, at which time the time-varying step size parameter is set to zero to stop the adaptive iteration of the filter: μ (m) ═ 0.
D. Normalizing the squared Euclidean norm of the error signal e (n) relative to the far-end signal u (n) of the loudspeaker to obtain a normalized error signal enor mAnd (n) smoothing the maximum value of each frame of the normalized error signal by using a convex combination first-order recursion process to be used as an upper limit threshold value of the normalized adaptive error.
Preferably, the step D specifically includes:
d1, squaring U (n) of the Euclidean norm of the current error signal e (n) relative to the speaker far-end signal U (n)2Normalization is carried out to obtain a normalized error signal enor m(n)：
Wherein the constant is a very small constant, which is used to prevent the denominator from being too small to cause divergence.
D2, obtaining absolute value | e of normalized error signal of each framenor mMaximum value e of (n) |max(m) smoothing said data with a convex combination first order recursive process to obtain a normalized error mean T (m) of the normalized error signal:
wherein, alpha and beta are forgetting factors, and the value of alpha is generally smaller: preferably, alpha is more than 0.001 and less than or equal to 0.005, beta is larger, and beta is more than 0.5 and less than 1;
d3, using the normalized error mean T (m) as the normalized error signal enor m(n) upper threshold to prevent error signal abrupt changes due to double-ended sounding and other interference conditions:
E. using the corrected time-varying step size parameter mu (m), normalizing the error signal enor m(n) iteratively updating FIR filter coefficients: w (n) ═ W (n-1) + μ (m) · enorm(n)·U(n)。
The present invention and the prior art are implemented below in conjunction with the HA320D digital hearing aid chip of Nanjing Tianyue electronics, Inc., where the HA320D digital hearing aid system sample rate is 16kHZ, and the quantization bit number is 16 bits; th1 ═ 0.5, Th2 ═ 0.7, Th4 ═ 0.05, Th3 needs to look at parameters according to system quantization under howling critical stateCondition measurement Pmax(m) to obtain; mu.saThe value is 0.05, mubThe value is 0.1; the value of alpha is 0.001, and the value of beta is 0.5; the value is 0.001.
As shown in fig. 2, in a real hearing aid closed-loop environment, the adaptive echo cancellation device and the echo cancellation method for a hearing aid in variable step sizes of the present invention have the following significant advantages:
1) the anti-interference capability under the external noise or double-end talkback scene is strong, and the iterative divergence of the adaptive filter can be obviously prevented;
2) imbalance can be effectively prevented when the device is in an environment with strong noise single-frequency energy;
3) the convergence speed and the steady-state error of the adaptive filter are well balanced, and a time-varying system and howling suppression can be quickly tracked;
4) by means of the multiplying unit, FFT and other modules of the hearing aid chip, the algorithm is low in complexity and power consumption.
The above description is only a preferred embodiment of the present invention, and not intended to limit the scope of the present invention, and all modifications of equivalent structures and equivalent processes, which are made by using the contents of the present specification and the accompanying drawings, or directly or indirectly applied to other related technical fields, are included in the scope of the present invention.
Claims (8)
1. A step size-variable hearing aid adaptive echo cancellation device is characterized by comprising a single-frequency tone detector, a step size controller and an adaptive filter:
the single-frequency tone detector is used for carrying out spectrum energy analysis on an error signal e (n) sample, calculating a current hearing aid system state parameter and transmitting the parameter to the step length controller, wherein the system state parameter comprises maximum single-frequency energy Pmax(m) and the ratio (m) of the maximum single-frequency energy to the total energy of the power spectrum;
the step length controller is used for judging the state of the system according to the system state parameter obtained from the single-frequency tone detector and the normalization correction coefficient f (m), calculating a time-varying step length parameter mu (m) of the self-adaptive filter, and transmitting the time-varying step length parameter mu (m) to the self-adaptive filter, specifically:
1) if the ratio (m) of the maximum single-frequency energy to the total energy of the power spectrum is smaller than a set threshold Th2, the system is judged to be in steady state convergence, and a preset steady state step length parameter mu is adoptedaAs a time-varying step size parameter;
2) if the ratio (m) of the maximum single-frequency energy to the total energy of the power spectrum is greater than a set threshold Th2, and the maximum single-frequency energy Pmax(m) is less than a set threshold Th3, the system is determined to be in a relevant interference state, and the time-varying step parameter is corrected by using a normalized correction coefficient f (m): μ (m) ═ f (m) · μa；
3) If the ratio (m) of the maximum single-frequency energy to the total energy of the power spectrum is greater than a set threshold Th2, and the maximum single-frequency energy Pmax(m) is greater than a set threshold Th3, and the normalization correction coefficient f (m) is less than a set threshold Th4, the system is determined to be in a howling state, at the moment, howling is preferentially processed, and a fixed step size parameter mu is adoptedbFilter iteration is accelerated: μ (m) ═ μbWherein, mub＞μa；
4) If the ratio (m) of the maximum single-frequency energy to the total energy of the power spectrum is greater than a set threshold Th2, and the maximum single-frequency energy Pmax(m) is greater than a set threshold Th3, and the normalized correction factor f (m) is greater than a set threshold Th4, then the system is determined to be in a double-ended sounding state, at which time the time-varying step size parameter is set to zero to stop the adaptive iteration of the filter: μ (m) ═ 0;
the adaptive filter is used for filtering the buffered loudspeaker far-end signal u (n) samples and calculating an estimated echo signal
Then outputting, and carrying out iterative updating on the self-adaptive filter according to the time-varying step length parameter calculated by the step length controller;
wherein the error signal e (n) is the near-end speech signal v (n) minus the estimated echo signal
2. The variable step size hearing aid adaptive echo cancellation device according to claim 1, wherein the single frequency tone detector comprises a frame FFT module and a spectral energy analysis module:
the frame-dividing FFT module is used for receiving a cached error signal e (n), carrying out fast Fourier transform on the error signal e (n) according to a set frequency, calculating a power spectrum of the signal and transmitting the power spectrum to the spectrum energy analysis module;
the spectrum energy analysis module analyzes the maximum single-frequency energy P of the obtained signal spectrummax(m) the ratio (m) of the maximum single-frequency energy to the total energy of the power spectrum, and transmitting the obtained system state parameters to the step length controller.
3. The adaptive echo cancellation device of claim 1, wherein the step size controller comprises a state selection module, and the state selection module selects a state of the system according to the system state parameters obtained by the single-tone detector, controls to adjust the time-varying step size parameter μ (m), and transmits the time-varying step size parameter μ (m) to the adaptive filter.
4. The variable step size hearing aid adaptive echo cancellation device according to claim 1, wherein the adaptive filter comprises a normalized error signal processing module, a coefficient update module, and a FIR filter module:
the normalization error signal processing module normalizes the square of the Euclidean norm of the error signal e (n) relative to the loudspeaker far-end signal u (n) to obtain a normalization error signal
Wherein, U (n) ═ { U (n), U (n +1),.. U (n + D-1) } is the far-end signal vector corresponding to the time delay, D is the adaptive filter order and is a constant;
normalizing error signal e according to set frequency by convex combination first-order recursion processnorm(n) performing a snapshot smoothing process to obtain a normalized errorA difference mean value T (m);
the coefficient updating module is used for updating the coefficient according to the normalized error signal enorm(n), the normalized error mean value T (m) and the time-varying step size parameter mu (m) obtained by the step size controller, and the adaptive filter coefficient is subjected to iterative updating;
the FIR filter module carries out filtering processing on the far-end signal u (n) to obtain an estimated echo signal
And then outputting.
5. The variable-stride hearing-aid adaptive echo cancellation device of any one of claims 1-4, wherein the variable-stride hearing-aid adaptive echo cancellation device is adapted for use in a digital hearing-aid system.
6. A method for adaptive echo cancellation in a variable step size hearing aid, comprising the steps of:
A. according to the cached far-end signal sample, the estimated echo signal is obtained through filtering of a self-adaptive FIR filter
Wherein W (n) { W ═ Wn(1)，wn(2)，...，wn(D)}，wn(x) FIR filter coefficients, WH(n) is the conjugate transpose of W (n); and subtracting the estimated echo signal from the sampled near-end speech signal v (n)
Obtaining an error signal
U (n) { U (n), U (n +1),.. U (n + D-1) } is the far-end signal vector corresponding to the time delay, and D is the adaptive filter order；
B. Performing time-frequency transformation on the error signal, calculating the power spectrum of the error signal and analyzing the energy distribution characteristics to obtain system state parameters, wherein the system state parameters comprise maximum single-frequency energy Pmax(m) and the ratio (m) of the maximum single-frequency energy to the total energy of the power spectrum;
C. judging the state of the system according to the system state parameters, and controlling the self-adaptive time-varying step length parameter mu (m); specifically, the method comprises the following steps:
c1, according to the ratio (m) of the maximum single-frequency energy obtained by the single-frequency tone detector to the total energy of the power spectrum, smoothing and normalizing the correction coefficient f (m):
f(m+1)＝λ·f(m)+(1-λ)·I(m)；
wherein, λ is a forgetting factor and is a positive number smaller than 1;
c2 maximum single frequency energy Pmax(m), the ratio (m) of the maximum single-frequency energy to the total energy of the power spectrum, and a normalization correction coefficient f (m) are used for judging a threshold value, and a time-varying step length parameter is adjusted according to the judgment results of the three state parameters, specifically:
1) if the ratio (m) of the maximum single-frequency energy to the total energy of the power spectrum is smaller than a set threshold Th2, the system is judged to be in steady state convergence, and a preset steady state step length parameter mu is adoptedaAs a time-varying step size parameter;
2) if the ratio (m) of the maximum single-frequency energy to the total energy of the power spectrum is greater than a set threshold Th2, and the maximum single-frequency energy Pmax(m) is less than a set threshold Th3, the system is determined to be in a relevant interference state, and the time-varying step parameter is corrected by using a normalized correction coefficient f (m): μ (m) ═ f (m) · μa；
3) If the ratio (m) of the maximum single-frequency energy to the total energy of the power spectrum is greater than a set threshold Th2, and the maximum single-frequency energy Pmax(m) is greater than a set threshold value Th3 and the normalized correction factor f (m) is less than a set threshold value Th4, the system is determined to be in a howling state, which isThe time first handles the howling, and adopts the fixed step length parameter mubFilter iteration is accelerated: μ (m) ═ μbWherein, mub＞μa；
4) If the ratio (m) of the maximum single-frequency energy to the total energy of the power spectrum is greater than a set threshold Th2, and the maximum single-frequency energy Pmax(m) is greater than a set threshold Th3, and the normalized correction factor f (m) is greater than a set threshold Th4, then the system is determined to be in a double-ended sounding state, at which time the time-varying step size parameter is set to zero to stop the adaptive iteration of the filter: μ (m) ═ 0;
D. normalizing the squared Euclidean norm of the error signal e (n) relative to the far-end signal u (n) of the loudspeaker to obtain a normalized error signal enorm(n), smoothing the maximum value of each frame of the normalized error signal by using a convex combination first-order recursion process, and taking the maximum value as an upper limit threshold of the normalized adaptive error;
E. using the corrected time-varying step size parameter mu (m), normalizing the error signal enorm(n) iteratively updating FIR filter coefficients: w (n) ═ W (n-1) + μ (m) · enorm(n)·U(n)。
7. The method according to claim 6, wherein the step B comprises:
b1, continuously filling error signals into a buffer area of the single-frequency tone detector, performing one-time fast Fourier transform operation after the buffer area is full of data, resetting the buffer to obtain the frequency spectrum of the signals, and further calculating the power spectrum PSD of the signalsm(k)：
PSDm(k)＝|fft(E(m))|2Wherein, m is the frame number, k is the frequency point, E (m) ═ { E (N), E (N +1),.., E (N + N-1) }; FFT (—) is the FFT operation; n is the number of fast Fourier transform points;
b2, searching the maximum energy frequency point k according to the power spectrum of the signalmaxCalculating the energy sum P of the maximum energy frequency point and the adjacent frequency pointsmax(m)：
Pmax(m)＝PSDm(kmax-1)+PSDm(kmax)+PSDm(kmax+1)；
Calculating the ratio (m) of the maximum single-frequency energy to the total energy of the power spectrum:
wherein N is the number of fast Fourier transform points;
a state parameter Pmax(m) and (m) are transmitted to the step size controller.
8. The method according to claim 6, wherein the step D specifically comprises:
d1, squaring U (n) of the Euclidean norm of the current error signal e (n) relative to the speaker far-end signal U (n)2Normalization is carried out to obtain a normalized error signal enorm(n)：
Wherein, is a constant;
d2, obtaining absolute value | e of normalized error signal of each framenormMaximum value e of (n) |max(m) smoothing said data with a convex combination first order recursive process to obtain a normalized error mean T (m) of the normalized error signal:
wherein, alpha and beta are forgetting factors;
d3, using the normalized error mean T (m) as the normalized error signal enorm(n) upper threshold limit.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011090523.1A CN111916099B (en) | 2020-10-13 | 2020-10-13 | Adaptive echo cancellation device and method for variable-step hearing aid |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011090523.1A CN111916099B (en) | 2020-10-13 | 2020-10-13 | Adaptive echo cancellation device and method for variable-step hearing aid |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111916099A CN111916099A (en) | 2020-11-10 |
| CN111916099B true CN111916099B (en) | 2020-12-29 |
Family
ID=73265222
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011090523.1A Active CN111916099B (en) | 2020-10-13 | 2020-10-13 | Adaptive echo cancellation device and method for variable-step hearing aid |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111916099B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114071220B (en) * | 2021-11-04 | 2024-01-19 | 深圳Tcl新技术有限公司 | Sound effect adjusting method and device, storage medium and electronic equipment |
| CN114584909B (en) * | 2022-04-29 | 2022-07-26 | 南京天悦电子科技有限公司 | Digital hearing aid howling suppression system and suppression method thereof |
Family Cites Families (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN100543842C (en) * | 2006-05-23 | 2009-09-23 | 中兴通讯股份有限公司 | Realize the method that ground unrest suppresses based on multiple statistics model and least mean-square error |
| CN101179294B (en) * | 2006-11-09 | 2012-07-04 | 黄山好视达通信技术有限公司 | Self-adaptive echo eliminator and echo eliminating method thereof |
| CN101320996B (en) * | 2008-05-27 | 2012-10-10 | 中山大学 | Self-adapting noise elimination apparatus and method |
| TWI424696B (en) * | 2010-11-29 | 2014-01-21 | Realtek Semiconductor Corp | Communication system and method for cancelling timing relevance of signal |
| US9344579B2 (en) * | 2014-07-02 | 2016-05-17 | Microsoft Technology Licensing, Llc | Variable step size echo cancellation with accounting for instantaneous interference |
| CN106782593B (en) * | 2017-02-27 | 2019-10-25 | 重庆邮电大学 | A kind of more band structure sef-adapting filter switching methods eliminated for acoustic echo |
| US10154148B1 (en) * | 2017-08-03 | 2018-12-11 | Polycom, Inc. | Audio echo cancellation with robust double-talk detection in a conferencing environment |
| CN108172233B (en) * | 2017-12-12 | 2019-08-13 | 天格科技（杭州）有限公司 | The echo cancel method of signal and error signal regression vectors is estimated based on distal end |
| CN109754813B (en) * | 2019-03-26 | 2020-08-25 | 南京时保联信息科技有限公司 | Variable step size echo cancellation method based on rapid convergence characteristic |
| CN111028856B (en) * | 2020-01-08 | 2022-01-28 | 西南交通大学 | Echo cancellation method with variable step length |
-
2020
- 2020-10-13 CN CN202011090523.1A patent/CN111916099B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN111916099A (en) | 2020-11-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| EP1228665B1 (en) | Feedback cancellation apparatus and methods utilizing an adaptive reference filter | |
| US6434246B1 (en) | Apparatus and methods for combining audio compression and feedback cancellation in a hearing aid | |
| Greenberg et al. | Evaluation of an adaptive beamforming method for hearing aids | |
| US8396234B2 (en) | Method for reducing noise in an input signal of a hearing device as well as a hearing device | |
| EP2201563B1 (en) | Multiple microphone voice activity detector | |
| EP3058710B1 (en) | Detecting nonlinear amplitude processing | |
| KR100423472B1 (en) | Gauging convergence of adaptive filters | |
| US20070230712A1 (en) | Telephony Device with Improved Noise Suppression | |
| EP2241099B1 (en) | Acoustic echo reduction | |
| CN112334972A (en) | Real-time detection of feedback instability | |
| KR20080059147A (en) | Robust separation of speech signals in a noisy environment | |
| EP1142288B1 (en) | Methods and apparatus for adaptive signal gain control in communications systems | |
| JP2012085314A (en) | Method and system for clear signal capture | |
| CN111916099B (en) | Adaptive echo cancellation device and method for variable-step hearing aid | |
| JP3402331B2 (en) | Noise reduction device | |
| EP3796680A1 (en) | Automatic timbre control | |
| WO2021016002A1 (en) | Frequency domain coefficient-based dynamic adaptation control of adaptive filter | |
| US10984778B2 (en) | Frequency domain adaptation with dynamic step size adjustment based on analysis of statistic of adaptive filter coefficient movement | |
| WO2021016001A1 (en) | Input signal-based frequency domain adaptive filter stability control | |
| CN112185404B (en) | Low-complexity double-end detection method based on subband signal-to-noise ratio estimation | |
| US11984107B2 (en) | Audio signal processing method and system for echo suppression using an MMSE-LSA estimator | |
| US20240021184A1 (en) | Audio signal processing method and system for echo supression using an mmse-lsa estimator | |
| DK1068773T4 (en) | Apparatus and method for combining audio compression and feedback suppression in a hearing aid | |
| CN113347536A (en) | Acoustic feedback suppression algorithm based on linear prediction and subband adaptive filtering |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |