Digital hearing aid howling suppression method and system

Abstract

The invention relates to the technical field of hearing aids, in particular to a digital hearing aid howling suppression method and system, and the method comprises the steps: determining an energy distribution feature value; based on the distribution of all high-energy values and derivative values of each to-be-processed signal and a preset number of to-be-processed signals before the to-be-processed signal in the frequency domain and the rate at which each peak value in each to-be-processed signal drops to an adjacent next valley value, determining the noise-containing possibility degree of each to-be-processed signal, and determining the noise-containing possibility degree of each to-be-processed signal in combination with the energy distribution characteristic value. Determining an over-subtraction index; and based on the over-subtraction index, suppressing the howling noise in the voice signal by adopting a spectral subtraction method. According to the method, the frequency domain energy distribution and the time domain oscillation characteristics are dynamically analyzed, the over-subtraction index of the spectral subtraction method is adaptively adjusted, the problem that a traditional howling suppression method is poor in adaptability is solved, and the howling suppression effect of the digital hearing aid in a complex acoustic environment is improved.

Description

Technical Field

[0001] This application relates to the field of hearing aid technology, specifically to a method and system for suppressing feedback in digital hearing aids. Background Technology

[0002] Digital hearing aids convert received speech signals from analog to digital. Through digital signal processing algorithms integrated into a digital signal processing chip, the speech signal is refined to improve its signal-to-noise ratio, thus assisting hearing. However, the speech signal output from the digital hearing aid's speaker is prone to leakage through gaps between the earpiece and the ear canal, or through internal vents, and can be picked up again by the microphone, creating acoustic feedback. This causes speech signal distortion, and in severe cases, complete distortion. When certain gain and phase conditions are met, the output speech signal can produce howling, affecting the performance of the digital hearing aid.

[0003] Currently, feedback suppression in digital hearing aids commonly employs a combination of notch filtering and spectral subtraction. Notch filtering suppresses feedback by attenuating the gain at the feedback frequency, while spectral subtraction aims to reduce the resulting speech distortion. However, this approach has significant drawbacks: the time-varying nature of speech signals leads to a random distribution of feedback frequencies, making it difficult for notch filtering to respond comprehensively; simultaneously, notch filtering itself introduces new time-varying noise, and existing spectral subtraction methods rely on fixed over-attenuation coefficients, failing to dynamically adjust according to the randomness and time-varying nature of feedback. This results in unsatisfactory suppression effects across different frequency bands, reducing the effectiveness of digital hearing aids in suppressing feedback noise. Summary of the Invention

[0004] To address the aforementioned technical problems, the purpose of this application is to provide a method and system for suppressing feedback in digital hearing aids, the specific technical solution of which is as follows: In a first aspect, embodiments of this application provide a method for suppressing feedback in a digital hearing aid, the method comprising the following steps: Acquire speech signals from a digital hearing aid and segment the speech signals into multiple signals to be processed; Based on the energy distribution of each signal to be processed in the frequency domain, the energy value of each signal to be processed in the frequency domain is divided into high and low energy values; based on the difference in the distribution between high and low energy values ​​of each signal to be processed in the frequency domain, and the degree of disorder of all energy values, the energy distribution characteristic value of each signal to be processed is determined. Based on the distribution of all high-energy values ​​and their derived values ​​in the frequency domain for each signal to be processed and a preset number of signals to be processed before it, a forward evaluation index for each signal to be processed is determined; based on the rate at which each peak in each signal to be processed drops to its next adjacent valley, an oscillation evaluation parameter for each signal to be processed is determined; based on the difference in oscillation evaluation parameters between each signal to be processed and a preset number of signals to be processed after it, a backward evaluation index for each signal to be processed is determined, and combined with the forward evaluation index, the noise probability of each signal to be processed is determined; Based on the energy distribution characteristic value and the noise probability, the over-attenuation exponent for each signal to be processed is determined; based on the over-attenuation exponent, spectral subtraction is used to suppress howling noise in the speech signal.

[0005] Preferably, dividing the energy value of each signal to be processed in the frequency domain into high and low energy values ​​includes: Calculate the mean of all energy values ​​of each signal to be processed in the frequency domain, and denote it as the global threshold. All energy values ​​of each signal to be processed that are greater than the global threshold are taken as high energy values, and the remaining energy values ​​are taken as low energy values.

[0006] Preferably, the method for determining the energy distribution characteristic value of each signal to be processed is as follows: Calculate the proportion of high energy values ​​with frequencies greater than the fundamental frequency in the frequency domain for each signal to be processed to the total number of high energy values ​​in the signal to be processed, and denote it as the high energy distribution value; calculate the proportion of low energy values ​​with frequencies greater than the fundamental frequency in the frequency domain for each signal to be processed to the total number of low energy values ​​in the signal to be processed, and denote it as the low energy distribution value; The difference between the high-energy distribution value and the low-energy distribution value is calculated. The information entropy of all energy in the frequency domain of each signal to be processed is positively fused with the difference, and the result is used as the energy distribution feature value of each signal to be processed.

[0007] Preferably, the method for determining the forward evaluation index of each signal to be processed is as follows: For each signal to be processed, obtain the mean, sum, maximum value, and number of all high-energy values ​​in the frequency domain; For each signal to be processed and the mean, sum of high energy values, maximum value, and number of high energy values ​​of a preset number of signals to be processed, four fitted lines are obtained. The sum of the slopes of the four fitted lines and the sum of the fitting errors are calculated respectively. The forward evaluation index of each signal to be processed is negatively correlated with the slope and value of its corresponding fitted line and the fitting error of the fitted line.

[0008] Preferably, the method for determining the oscillation evaluation parameter of each signal to be processed is as follows: The rate at which each peak in each signal to be processed drops to its next adjacent trough is denoted as the rate of decline of each peak. The average of the differences in the rates of decline between all peaks and their next adjacent peaks is used as the oscillation evaluation parameter for each signal to be processed.

[0009] Preferably, the backward evaluation index for each signal to be processed is the sum of the differences in oscillation evaluation parameters between each signal to be processed and a preset number of subsequent signals to be processed.

[0010] Preferably, the noise probability of each signal to be processed is the result of the forward fusion of the normalized value of the forward evaluation index and the normalized value of the backward evaluation index of each signal to be processed.

[0011] Preferably, the expression for the over-decrease exponent of each signal to be processed is: In the formula, , These represent the over-attenuation exponents of the i-th and (i-1)-th signals to be processed, respectively, where the initial value of the over-attenuation exponent is a preset value; , Let represent the spectral smoothing coefficients of the i-th and (i-1)-th signals to be processed, respectively. The spectral smoothing coefficient of each signal to be processed is the normalized value of the energy distribution characteristic value and the normalized mean of the noise probability. This indicates the preset adjustment step size.

[0012] Preferably, the method of suppressing howling noise in the speech signal using spectral subtraction includes: The estimated noise spectrum of each frame of the signal to be processed is obtained. The spectra of all signals to be processed and their estimated noise spectra are used as inputs to spectral subtraction. The over-subtraction exponents of all signals to be processed are used as over-detection factors in spectral subtraction. The denoised signals to be processed are then output. Secondly, embodiments of this application also provide a digital hearing aid feedback suppression system, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of any one of the above-described digital hearing aid feedback suppression methods.

[0013] This application has at least the following beneficial effects: This application innovatively constructs energy distribution feature values ​​by analyzing the energy distribution characteristics of the signal to be processed in the frequency domain and combining information entropy with the difference between high and low energy distribution values. This accurately identifies the typical characteristics of residual feedback noise, thereby effectively distinguishing noise from valid speech and helping to improve the feedback suppression effect of digital hearing aids in complex acoustic environments. Furthermore, this application constructs forward and backward bidirectional evaluation indicators by fusing time-domain oscillation characteristics and frequency-domain energy attenuation characteristics, achieving accurate identification and quantitative evaluation of residual feedback noise, which helps to improve the feedback suppression effect of digital hearing aids. Finally, by fusing energy distribution feature values ​​and noise probability to construct a spectral subtraction smoothing coefficient, and adaptively adjusting the over-subtraction exponent based on the dynamic changes of this coefficient between adjacent frames, real-time optimization of spectral subtraction noise suppression intensity is achieved. This effectively suppresses residual feedback noise while maximizing the preservation of the original quality of the speech signal, thus improving the feedback suppression effect of digital hearing aids in complex acoustic environments. Attached Figure Description

[0014] To more clearly illustrate the technical solutions and advantages in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 A flowchart illustrating the steps of a digital hearing aid feedback suppression method according to one embodiment of this application; Figure 2 This is a schematic diagram of the spectral subtraction smoothing coefficient extraction process provided in one embodiment of this application. Detailed Implementation

[0016] To further illustrate the technical means and effects adopted by this application to achieve the intended purpose of the invention, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a digital hearing aid feedback suppression method and system proposed in this application. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0017] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0018] The following description, in conjunction with the accompanying drawings, details a specific scheme for a digital hearing aid feedback suppression method and system provided in this application.

[0019] Please see Figure 1 The diagram illustrates a flowchart of a digital hearing aid feedback suppression method according to an embodiment of this application, the method comprising the following steps: Step S1: Acquire the speech signal from the digital hearing aid.

[0020] In this embodiment, the sampling frequency of the digital hearing aid's speech signal is 44.1kHz. To facilitate data processing, the speech signal is divided into frames, with the frame length of each speech frame set to 25ms. At the same time, to reduce distortion during speech signal processing, this embodiment adopts an alternating framing method, where adjacent speech frames overlap, and the frame shift is set to 1 / 2 of the speech frame length.

[0021] To reduce spectral leakage and boundary effects caused by audio signal framing, this embodiment uses a window function to window each audio frame.

[0022] Furthermore, since the cause of feedback is the sound amplified by the hearing aid being transmitted to the microphone through spatial or structural vibrations, and then becoming an input signal again, which is cyclically amplified and generates audio resonance, this embodiment first uses a notch filter to suppress acoustic feedback for each speech frame. Specifically, this embodiment uses an adaptive notch filter (ANF) to suppress acoustic feedback for each speech frame, and each speech frame after acoustic feedback suppression is denoted as a signal to be processed.

[0023] Step S2: Based on the energy distribution of each signal to be processed in the frequency domain, divide the energy value of each signal to be processed in the frequency domain into high and low energy values; based on the difference in the distribution between high and low energy values ​​of each signal to be processed in the frequency domain, and the degree of disorder of all energy values, determine the energy distribution characteristic value of each signal to be processed.

[0024] The cause of howling lies in the acoustic feedback loop: the sound output by the hearing aid speaker leaks to the microphone through the air or structural vibration path, forming a closed loop. When the gain and phase of the loop meet the oscillation condition, the signal is amplified in a loop, eventually triggering self-excited oscillation at a specific frequency, which is howling. Howling can be divided into internal howling and external howling according to its cause. In this embodiment, spectral subtraction is used to suppress howling. The effectiveness of spectral subtraction is based on a specific premise: that is, the target noise is unrelated to the speech signal and its spectral characteristics can be statistically analyzed or estimated.

[0025] Since the signal to be processed obtained in step S1 is a superposition of the effective speech signal and the residual noise introduced by acoustic feedback suppression, compared with the effective speech signal, the residual noise exhibits stronger randomness and time-varying characteristics in the frequency domain. Its energy distribution varies significantly across different speech frames and frequency points. Based on this characteristic, this embodiment divides the energy value of each signal to be processed into high and low energy values ​​in the frequency domain based on the energy distribution of each signal to be processed. Based on the differences in the distribution between high and low energy values ​​in the frequency domain and the degree of disorder of all energy values, the energy distribution characteristic value of each signal to be processed is determined to accurately identify and filter out the frequency domain regions contaminated by noise. Specifically: First, this embodiment divides the energy value of each signal to be processed into high and low energy values ​​based on the energy distribution of each signal in the frequency domain. Specifically: In this embodiment, the mean of all energy values ​​of each signal to be processed in the frequency domain is calculated and denoted as the global threshold. All energy values ​​of each signal to be processed that are greater than the global threshold are taken as high energy values, and the remaining energy values ​​are taken as low energy values.

[0026] It should be noted that the energy value of each signal to be processed in the frequency domain can be obtained by using the short-time Fourier transform algorithm to obtain the time-frequency graph of each signal. The horizontal axis of the time-frequency graph is time, and the vertical axis is frequency. The energy level of each frequency point is represented by the color depth in the two-dimensional time-frequency graph. Generally, the darker the color, the higher the energy. Therefore, the size of the color value can be used to represent the size of the energy value. In this embodiment, the color ranges from black to white as an example. Black (pixel value of 0) represents the minimum energy value, and white (pixel value of 255) represents the maximum energy value. It should be noted that in other embodiments, different colors can be set to represent different energy values. This embodiment does not impose any special restrictions.

[0027] The short-time Fourier transform algorithm is a well-known technique, and the specific process of obtaining the time-frequency diagram of the signal using it will not be elaborated here.

[0028] Furthermore, this embodiment determines the energy distribution characteristic value of each signal to be processed based on the difference in the distribution between high and low energy values ​​in the frequency domain, and the degree of disorder of all energy values. This characteristic value is used to characterize the energy value distribution in each frame of the signal to be processed. Specifically: In this embodiment, the proportion of the number of high energy values ​​with frequencies greater than the fundamental frequency in the frequency domain of each signal to be processed is calculated to the total number of high energy values ​​in the signal to be processed, and is denoted as the high energy distribution value; the proportion of the number of low energy values ​​with frequencies greater than the fundamental frequency in the frequency domain of each signal to be processed is calculated to the total number of low energy values ​​in the signal to be processed, and is denoted as the low energy distribution value. The difference between the high-energy distribution value and the low-energy distribution value is calculated. The information entropy of all energy in the frequency domain of each signal to be processed is positively fused with the difference, and the result is used as the energy distribution feature value of each signal to be processed.

[0029] The baseband frequency is a well-known technology, and its specific acquisition method will not be elaborated here.

[0030] It should be understood that positive fusion refers to combining two or more indicators through addition or multiplication to obtain a comprehensive indicator, thereby more comprehensively and accurately assessing a phenomenon or problem. This fusion method is not limited to simple arithmetic operations, but can also include more complex statistical models and analytical methods. Implementers can choose according to specific circumstances, and this embodiment does not impose any special restrictions.

[0031] Preferably, as one implementation method, in this embodiment, the product of the information entropy of all energy of each signal to be processed in the frequency domain and the difference is taken as the result of positive fusion of the information entropy of all energy of each signal to be processed in the frequency domain and the difference. In practical applications, as other implementation methods, implementers may also adopt other positive fusion methods such as addition or weighted fusion according to specific circumstances. This embodiment does not impose any special restrictions.

[0032] Based on the energy distribution characteristic value of each signal to be processed, it can be understood that the energy distribution characteristic value is used to characterize the noise distribution characteristics of the signal to be processed in the frequency domain. The information entropy of all energy in the frequency domain of the signal to be processed reflects the randomness of the energy distribution. The difference between the high energy distribution value and the low energy distribution value is the bias factor of the energy distribution in the high frequency region, reflecting the significance of high frequency noise. If the information entropy of all energy in the frequency domain of the current signal to be processed is larger, it indicates that the noise content in the current signal to be processed is higher. At the same time, if the difference between the high energy distribution value and the low energy distribution value of the current signal to be processed is smaller, it indicates that the proportion of high frequency energy in the current signal to be processed is higher. This is a typical characteristic value of howling residual noise, indicating that the possibility of howling residual noise in the current signal to be processed is greater. Therefore, the corresponding energy distribution noise is also greater. Conversely, if the information entropy of all energy in the current signal to be processed is smaller in the frequency domain, it indicates that the energy distribution in the current signal to be processed is more concentrated, the randomness is lower, the signal is closer to pure speech, and its noise content is also lower. At the same time, if the difference between the high energy distribution value and the low energy distribution value of the current signal to be processed is larger, it indicates that the energy distribution in the frequency domain of the current signal to be processed is more dispersed, and low-frequency energy occupies a dominant position. This does not conform to the characteristic of high-frequency concentration of howling residual noise, indicating that the possibility of howling residual noise in the current signal to be processed is smaller. Therefore, the corresponding energy distribution noise is also smaller.

[0033] Thus, this embodiment analyzes the energy distribution characteristics of the signal to be processed in the frequency domain and innovatively combines information entropy with the difference between high and low energy distribution values ​​to construct energy distribution feature values, so as to accurately identify the typical characteristics of residual noise of howling, thereby effectively distinguishing noise from effective speech, which helps to improve the howling suppression effect of digital hearing aids in complex acoustic environments.

[0034] Step S3: Based on the distribution of all high energy values ​​and their derived values ​​in the frequency domain for each signal to be processed and a preset number of signals to be processed before it, determine the forward evaluation index for each signal to be processed; based on the rate at which each peak in each signal to be processed drops to its next adjacent valley value, determine the oscillation evaluation parameter for each signal to be processed; based on the difference in the oscillation evaluation parameter between each signal to be processed and a preset number of signals to be processed after it, determine the backward evaluation index for each signal to be processed, and combine the forward evaluation index to determine the noise probability of each signal to be processed.

[0035] The essence of howling is the self-excited oscillation generated at a specific frequency point by the closed-loop system formed by the acoustic feedback path in the digital hearing aid. After the initial acoustic feedback suppression, the energy of the oscillation signal is significantly reduced, but the residual time-varying noise still exhibits characteristics that are completely different from the conventional speech signal in the time and frequency domain.

[0036] Specifically, in the frequency domain, howling manifests as energy abrupt changes and sharp resonance peaks. That is, at the instant howling is suppressed, the narrow-band frequency points whose energy was originally boosted due to self-excited oscillation will experience a sudden energy abrupt change. By comparing and analyzing the spectrum of the historical signal to be processed, it can be observed that the energy value of the howling frequency and its adjacent frequency bands drops sharply regardless of the frequency point. At the same time, due to the physical effect of acoustic resonance, the spectrum corresponding to howling will contain one or more sharp spectral peaks. This is a significant frequency domain feature that distinguishes howling signals from broadband noise and general speech signals. In addition, from the time domain perspective, howling manifests as damped oscillation and transient decay. That is, after the howling signal is suppressed, its energy does not instantly return to zero, but enters a damped oscillation process. This means that the amplitude envelope of multiple consecutive signals to be processed after the current signal will exhibit an exponentially decaying oscillation centered on the howling resonance frequency, which is fundamentally different from the natural oscillation and release process of speech signals.

[0037] Therefore, based on the above analysis, this embodiment determines the forward evaluation index of each signal to be processed based on the distribution of all high energy values ​​and their derived values ​​in the frequency domain of each signal to be processed and a preset number of signals to be processed before it; it determines the oscillation evaluation parameter of each signal to be processed based on the rate at which each peak in each signal to be processed drops to its adjacent trough; it determines the backward evaluation index of each signal to be processed based on the difference in the oscillation evaluation parameter between each signal to be processed and a preset number of signals to be processed after it; and, in combination with the forward evaluation index, it determines the noise probability of each signal to be processed in order to identify residual howling noise. The specific process is as follows: In this embodiment, firstly, based on the distribution of all high-energy values ​​and their derived values ​​in the frequency domain of each signal to be processed and a preset number of previous signals to be processed, a forward evaluation index for each signal to be processed is determined. Specifically: This embodiment calculates the mean and sum of all high-energy values ​​of each signal to be processed in the frequency domain, and counts the maximum value and the number of all high-energy values ​​of each signal to be processed in the frequency domain. Furthermore, the average high-energy value, the cumulative sum of high-energy values, the maximum high-energy value, and the number of high-energy values ​​of each signal to be processed and the previous preset number of signals to be processed are fitted to obtain four fitted lines. The sum of the slopes of the four fitted lines and the sum of the fitting errors of the four fitted lines are calculated respectively. The forward evaluation index of each signal to be processed is negatively correlated with the slope and value of its corresponding fitted line and the fitting error of the fitted line.

[0038] It should be noted that the preset number is set manually. In this embodiment, the preset number is 20, and the preset number of signals to be processed are adjacent to and continuous with each of the signals to be processed. In actual application, as other implementation methods, the implementer can also set the preset number according to the specific situation. This embodiment does not impose any special restrictions.

[0039] It should be noted that there are many commonly used fitting algorithms. In this embodiment, the least squares method is used to fit the mean of high energy values, the sum of high energy values, the maximum value of high energy values, and the number of high energy values. In practical applications, as other implementation methods, implementers may also use other fitting methods such as polynomial function fitting according to specific circumstances. This embodiment does not impose any special restrictions.

[0040] The methods for calculating the fitting error and the least squares method are well-known techniques. The calculation process of the fitting error and the specific process of fitting the data using the least squares method will not be described in detail.

[0041] It should be understood that a negative correlation means that the dependent variable decreases as the independent variable increases, and the dependent variable increases as the independent variable decreases. It can be a subtraction relationship or a division relationship, depending on the actual application.

[0042] Preferably, as one implementation method, in this embodiment, the expression for the forward evaluation index of each signal to be processed is: In the formula, This represents the forward evaluation index of the i-th signal to be processed; This represents the sum of the slopes of the four fitted lines corresponding to the i-th signal to be processed; This represents the sum of the fitting errors of the four fitted lines corresponding to the i-th signal to be processed; exp() represents an exponential function with the natural constant as the base. This indicates a preset constant greater than 0, used to prevent the denominator from being 0. The value is set manually, in this embodiment. The value of is 0.01. In practical applications, as other implementation methods, under the premise of ensuring that the denominator is not 0 and does not excessively affect the calculation result, the implementer can also set it according to the specific situation. This embodiment does not impose any special restrictions.

[0043] Based on the forward evaluation index of each signal to be processed, it can be understood that the forward evaluation index is used to characterize the decay trend of noise energy in the signal to be processed and its historical signals in time. The slope and sum of the four fitted lines corresponding to the signal to be processed reflect the decay rate of noise energy. The sum of the fitting errors of the four fitted lines corresponding to the signal to be processed reflects the linear consistency of energy fitting, that is, the stability and consistency of the energy decay process. If the slope and sum of the four fitted lines corresponding to the current signal to be processed are larger, it means that the energy decay of the current signal to be processed and the previous signals to be processed is faster, and the corresponding forward evaluation index is smaller, indicating that the noise of the current signal to be processed is decaying rapidly. At the same time, if the sum of the fitting errors of the four fitted lines corresponding to the current signal to be processed is smaller, it means that the energy decay process is very smooth and monotonous, without sudden changes in speed. This is another characteristic of the decay of residual howling noise. This means that the current signal to be processed is more likely to have howling, and the corresponding forward evaluation index is larger, indicating that the noise in the current signal to be processed is more likely to be residual howling noise. Conversely, if the sum of the slopes of the four fitted lines corresponding to the current signal to be processed is smaller, it indicates that the energy decay of the current signal to be processed and the previous signals to be processed is slower, or even tends to be stable or shows energy recovery. The larger the corresponding forward evaluation index, the more it indicates that the energy of the current signal to be processed does not show typical decay characteristics. At the same time, if the sum of the fitting errors of the four fitted lines corresponding to the current signal to be processed is larger, it indicates that the energy decay process fluctuates violently and has poor linear consistency. Its energy changes are chaotic and do not conform to the characteristics of smooth decay of residual howling noise. This indicates that the possibility of howling in the current signal to be processed is smaller, and the corresponding forward evaluation index is smaller, indicating that the noise in the current signal to be processed is more likely to be random background noise rather than residual howling noise.

[0044] Furthermore, this embodiment determines the oscillation evaluation parameters for each signal to be processed based on the rate at which each peak value in each signal to be processed decreases to its adjacent trough value. Specifically: In this embodiment, the rate at which each peak in each signal to be processed drops to its next adjacent trough is denoted as the rate of decline of each peak. The average of the differences in the rates of decline between all peaks and their next adjacent peaks is used as the oscillation evaluation parameter for each signal to be processed.

[0045] It should be noted that the calculation of the rate of decline is the result of dividing the difference between each peak value and its adjacent trough value by the time interval between them.

[0046] Based on the oscillation evaluation parameters of each signal to be processed, it can be understood that the oscillation evaluation parameters are used to characterize the damped oscillation characteristics of the signal to be processed in the time domain after the howling is suppressed. The greater the difference in the rate of decrease between all peaks of the signal to be processed and its next adjacent peak, the faster the oscillation amplitude decays over time, corresponding to the exponential decay characteristic after the howling is suppressed. The larger the corresponding oscillation evaluation parameters, the more significant the damped oscillation is in the current signal to be processed, i.e., the howling noise remains. Conversely, if the difference in the rate of decrease between all peaks of the current signal to be processed and their adjacent peaks is smaller, it indicates that the oscillation amplitude decays more slowly over time, or even has no obvious decay trend, and the amplitude remains stable or fluctuates randomly. This does not conform to the exponential decay characteristics that should be present after the howling is suppressed. The smaller the corresponding oscillation evaluation parameter, the less significant the damped oscillation in the current signal to be processed. That is, the signal is more likely to be a smooth speech or random noise, rather than a residual howling noise.

[0047] Furthermore, in this embodiment, based on the difference in oscillation evaluation parameters between each signal to be processed and a preset number of subsequent signals to be processed, a backward evaluation index for each signal to be processed is determined. Combined with the forward evaluation index, the noise probability of each signal to be processed is determined, specifically as follows: In this embodiment, the sum of the differences between the oscillation evaluation parameters of each signal to be processed and the preset number of signals to be processed after it is used as the backward evaluation index of each signal to be processed.

[0048] It should be noted that the preset number is set manually. In this embodiment, the preset number is 20, and the preset number of signals to be processed are adjacent to and continuous with each signal to be processed. In actual application, as other implementation methods, the implementer can also set the preset number according to the specific situation. This embodiment does not impose any special restrictions.

[0049] Based on the backward evaluation index of each signal to be processed, it can be understood that the backward evaluation index is used to characterize the attenuation consistency of the signal to be processed and its subsequent signals to be processed in terms of oscillation characteristics. If the difference between the current signal to be processed and the preset number of signals to be processed after it is larger, it indicates that the oscillation intensity of the current signal to be processed is significantly higher than that of the subsequent signals to be processed, which is in line with the physical law that the oscillation amplitude decays with time after the howling is suppressed. Therefore, the larger the corresponding backward evaluation index, the more it indicates that the current signal to be processed is the starting point of howling noise. Conversely, if the difference between the oscillation evaluation parameter of the current signal to be processed and the preset number of signals to be processed after it is smaller, it indicates that the oscillation intensity of the current signal to be processed is similar to or has no significant difference from the subsequent signals to be processed. The oscillation phenomenon does not show a trend of decay over time, which does not conform to the physical law that the oscillation gradually weakens after howling suppression. Therefore, the smaller the corresponding backward evaluation index, the more likely its oscillation characteristics are caused by a stable speech signal or continuous random background noise.

[0050] Furthermore, in this embodiment, the result of the forward evaluation index normalization value and the backward evaluation index normalization value of each signal to be processed is used as the noise probability of each signal to be processed.

[0051] Preferably, as one implementation method, in this embodiment, the sum of the normalized values ​​of the forward evaluation index and the normalized values ​​of the backward evaluation index of each signal to be processed is used as the noise probability of each signal to be processed. In practical applications, as other implementation methods, implementers may also adopt other forward fusion methods such as multiplication according to specific circumstances. This embodiment does not impose any special restrictions.

[0052] Based on the noise probability of each signal to be processed, it can be understood that the noise probability reflects the likelihood that the signal to be processed contains residual noise generated by the howling suppression process, and is used to characterize the noise pollution level of the current signal to be processed. If the forward evaluation index of the current signal to be processed is larger, it means that the probability of noise in the current signal to be processed is larger, and therefore, the corresponding noise probability is larger. At the same time, if the backward evaluation index of the current signal to be processed is larger, it means that the current signal to be processed has triggered strong time-domain oscillations, and these oscillations show a significant attenuation trend in subsequent signals to be processed. This is the typical oscillation attenuation characteristic left in the time domain after howling is generated by resonance and then suppressed. This means that the probability of howling residual noise in the current signal to be processed is larger, and therefore, the corresponding noise probability is larger. Conversely, the smaller the forward evaluation index of the current signal to be processed, the lower the probability of noise in the current signal to be processed, and therefore the lower the corresponding noise level. At the same time, the smaller the backward evaluation index of the current signal to be processed, the lower the probability that the current signal to be processed has not triggered strong time-domain oscillation, or that the oscillation phenomenon has not shown an obvious attenuation trend in the subsequent signal to be processed, which does not meet the oscillation attenuation characteristics that should be present after the howling is suppressed. This means that the probability that there is residual howling noise in the current signal to be processed is lower, and therefore the corresponding noise level is lower.

[0053] Thus, this embodiment constructs a forward and backward bidirectional evaluation index by integrating time-domain oscillation characteristics and frequency-domain energy attenuation characteristics, achieving accurate identification and quantitative evaluation of residual feedback noise, which helps to improve the feedback suppression effect of digital hearing aids.

[0054] Step S4: Based on the energy distribution characteristic value and the noise probability, determine the over-attenuation exponent for each signal to be processed; based on the over-attenuation exponent, use spectral subtraction to suppress howling noise in the speech signal.

[0055] Based on the energy distribution characteristic value obtained in step S2 and the noise probability obtained in step S3, the over-subtraction exponent for each signal to be processed is determined. Furthermore, based on the over-subtraction exponent, spectral subtraction is used to suppress howling noise in the speech signal, specifically as follows: As one implementation method, in this embodiment, the over-decrease exponent of the i-th signal to be processed is... The expression is: In the formula, This represents the over-decrease exponent of the (i-1)th signal to be processed, where the initial value of the over-decrease exponent is a preset value; , Let represent the spectral smoothing coefficients of the i-th and (i-1)-th signals to be processed, respectively. The spectral smoothing coefficient of each signal to be processed is the normalized value of the energy distribution characteristic value and the normalized mean of the noise probability. This indicates the preset adjustment step size.

[0056] Preferably, the spectral subtraction smoothing coefficient extraction process provided in this embodiment is illustrated as follows: Figure 2 As shown.

[0057] It should be noted that the preset adjustment step size is set manually, and its value range is generally [0.5, 1]. In this embodiment, the preset adjustment step size is 0.5. In actual application, as other implementation methods, implementers can also set it according to specific circumstances. This embodiment does not impose any special restrictions.

[0058] It should be noted that the preset value is also set manually. In this embodiment, the preset value is 1. In actual application, as other implementation methods, implementers can also set it according to specific circumstances. This embodiment does not impose any special restrictions.

[0059] Based on the over-subtraction exponent for each signal to be processed, it can be understood that the over-subtraction exponent is used to dynamically adjust the noise suppression strength of spectral subtraction. The spectral subtraction smoothing coefficient reflects the noise significance of the frequency domain energy distribution and time domain oscillation characteristics. If the difference between the spectral subtraction smoothing coefficient of the current signal to be processed and its adjacent previous signal to be processed is larger, it indicates that the noise content in the current signal to be processed is higher, the high-frequency energy is concentrated, and the oscillation attenuation is significant. Therefore, the corresponding over-subtraction exponent increases, enhancing the denoising strength of spectral subtraction. Conversely, if the difference between the spectral subtraction smoothing coefficient of the current signal to be processed and its adjacent previous signal to be processed is smaller, it indicates that the noise content in the current signal to be processed is lower, the frequency domain energy distribution is more stable, the time domain oscillation characteristics are not significant, and the signal is closer to clean speech. Therefore, the corresponding over-subtraction exponent decreases, reducing the denoising strength of spectral subtraction to avoid unnecessary damage to the effective speech signal.

[0060] Furthermore, in this embodiment, the estimated noise spectrum of each frame of the signal to be processed is obtained, the spectrum of all the signals to be processed and their estimated noise spectrum are used as inputs to spectral subtraction, the over-subtraction exponents of all the signals to be processed are used as over-detection factors in spectral subtraction, and the denoised signals to be processed are output.

[0061] It should be noted that noise spectrum estimation and spectral subtraction are well-known techniques, and the specific process of noise spectrum estimation and spectral subtraction for suppressing howling will not be elaborated here.

[0062] Thus, this embodiment constructs a spectral subtraction smoothing coefficient by fusing energy distribution characteristic values ​​with noise probability, and adaptively adjusts the over-subtraction exponent based on the dynamic changes of this coefficient between adjacent frames, thereby achieving real-time optimization of the spectral subtraction noise suppression intensity. This effectively suppresses residual feedback noise while preserving the original quality of the speech signal to the maximum extent, significantly improving the feedback suppression effect of digital hearing aids in complex acoustic environments.

[0063] Based on the same inventive concept as the above methods, this application also provides a digital hearing aid feedback suppression system, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of any one of the above-described digital hearing aid feedback suppression methods.

[0064] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, specific embodiments of this specification have been described above. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0065] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0066] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the principles of this application should be included within the protection scope of this application.

Claims

1. A method for suppressing whistling in a digital hearing aid, characterized in that, The method includes the following steps: Acquire speech signals from a digital hearing aid and segment the speech signals into multiple signals to be processed; Based on the energy distribution of each signal to be processed in the frequency domain, the energy value of each signal to be processed in the frequency domain is divided into high and low energy values; based on the difference in the distribution between high and low energy values ​​of each signal to be processed in the frequency domain, and the degree of disorder of all energy values, the energy distribution characteristic value of each signal to be processed is determined. Based on the distribution of all high-energy values ​​and their derived values ​​in the frequency domain for each signal to be processed and a preset number of signals to be processed before it, a forward evaluation index for each signal to be processed is determined; based on the rate at which each peak in each signal to be processed drops to its next adjacent valley, an oscillation evaluation parameter for each signal to be processed is determined; based on the difference in oscillation evaluation parameters between each signal to be processed and a preset number of signals to be processed after it, a backward evaluation index for each signal to be processed is determined, and combined with the forward evaluation index, the noise probability of each signal to be processed is determined; Based on the energy distribution characteristic value and the noise probability, the over-attenuation exponent for each signal to be processed is determined; based on the over-attenuation exponent, spectral subtraction is used to suppress howling noise in the speech signal.

2. The method for suppressing feedback in a digital hearing aid as described in claim 1, characterized in that, The step of dividing the energy value of each signal to be processed in the frequency domain into high and low energy values ​​includes: Calculate the mean of all energy values ​​of each signal to be processed in the frequency domain, and denote it as the global threshold. All energy values ​​of each signal to be processed that are greater than the global threshold are taken as high energy values, and the remaining energy values ​​are taken as low energy values.

3. The method for suppressing feedback in a digital hearing aid as described in claim 1, characterized in that, The method for determining the energy distribution characteristic value of each signal to be processed is as follows: Calculate the proportion of the number of high energy values ​​with frequencies greater than the fundamental frequency in the frequency domain of each signal to be processed to the total number of high energy values ​​in the signal to be processed, and denote it as the high energy distribution value. Calculate the proportion of the number of low energy values ​​with frequencies greater than the fundamental frequency in the frequency domain of each signal to be processed to the total number of low energy values ​​in the signal to be processed, and denote it as the low energy distribution value. The difference between the high-energy distribution value and the low-energy distribution value is calculated. The information entropy of all energy in the frequency domain of each signal to be processed is positively fused with the difference, and the result is used as the energy distribution feature value of each signal to be processed.

4. The method for suppressing feedback in a digital hearing aid as described in claim 1, characterized in that, The method for determining the forward evaluation index of each signal to be processed is as follows: For each signal to be processed, obtain the mean, sum, maximum value, and number of all high-energy values ​​in the frequency domain; For each signal to be processed and the mean, sum of high energy values, maximum value, and number of high energy values ​​of a preset number of signals to be processed, four fitted lines are obtained. The sum of the slopes of the four fitted lines and the sum of the fitting errors are calculated respectively. The forward evaluation index of each signal to be processed is negatively correlated with the slope and value of its corresponding fitted line and the fitting error of the fitted line.

5. The method for suppressing feedback in a digital hearing aid as described in claim 1, characterized in that, The method for determining the oscillation evaluation parameters for each signal to be processed is as follows: The rate at which each peak in each signal to be processed drops to its next adjacent trough is denoted as the rate of decline of each peak. The average of the differences in the rates of decline between all peaks and their next adjacent peaks is used as the oscillation evaluation parameter for each signal to be processed.

6. The method for suppressing feedback in a digital hearing aid as described in claim 1, characterized in that, The backward evaluation index for each signal to be processed is the sum of the differences in oscillation evaluation parameters between each signal to be processed and a preset number of subsequent signals to be processed.

7. The method for suppressing feedback in a digital hearing aid as described in claim 1, characterized in that, The noise probability of each signal to be processed is the result of the positive fusion of the normalized value of the forward evaluation index and the normalized value of the backward evaluation index of each signal to be processed.

8. The method for suppressing feedback in a digital hearing aid as described in claim 1, characterized in that, The expression for the over-decrease exponent of each signal to be processed is: In the formula, , These represent the over-attenuation exponents of the i-th and (i-1)-th signals to be processed, respectively, where the initial value of the over-attenuation exponent is a preset value; , Let represent the spectral smoothing coefficients of the i-th and (i-1)-th signals to be processed, respectively. The spectral smoothing coefficient of each signal to be processed is the normalized value of the energy distribution characteristic value and the normalized mean of the noise probability. This indicates the preset adjustment step size.

9. A method for suppressing whistling in a digital hearing aid as described in claim 1, characterized in that, The method of suppressing howling noise in speech signals using spectral subtraction includes: The estimated noise spectrum of each frame of the signal to be processed is obtained. The spectra of all the signals to be processed and their estimated noise spectra are used as inputs to the spectral subtraction method. The over-subtraction exponents of all the signals to be processed are used as over-detection factors in the spectral subtraction method. The denoised signals to be processed are then output.

10. A digital hearing aid feedback suppression system, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the digital hearing aid feedback suppression method as described in any one of claims 1-9.

Images