Digital hearing aid automatic gain control method and system

Abstract

The invention relates to the technical field of hearing aids, in particular to an automatic gain control method and system for a digital hearing aid, and the method comprises the steps: collecting a sound signal through a built-in microphone of the hearing aid; obtaining a Mel-language spectrogram of the sound signal; equally dividing each frame of Mel-language spectrogram into Mel-frequency bands; determining the energy concentration ratio of each Mel frequency band; determining the noise saliency of each frame of Mel-language spectrogram; blind source separation is carried out on the collected sound signals by adopting a natural gradient algorithm, the separation difference degree of each iteration is obtained through the similarity degree of the target signals obtained after each iteration separation and the noise signals and the similarity degree of the collected sound signals and the separated signals in combination with the noise saliency of all frames of Mel-language spectrograms, and the separation difference degree of each iteration is obtained; determining the step length of the next iteration; and performing automatic gain control on a target signal obtained by blind source separation. Therefore, the frequency interference of the hearing aid is reduced, and the hearing experience of the user is optimized.

Description

Technical Field

[0001] This application relates to the field of hearing aid technology, specifically to a method and system for automatic gain control of a digital hearing aid. Background Technology

[0002] Hearing aids are effective devices that improve the hearing level of hearing-impaired patients, helping them to hear clear and complete sounds from the outside world. With the development of digital signal processing technology, digital hearing aids, with their excellent sound signal processing capabilities, have built-in automatic gain control methods to process the collected sound signals by noise reduction, compression, compensation, and direction positioning, reducing environmental noise and enabling wearers to accurately identify and enhance the required sound signals, thereby improving the hearing ability and comfort of hearing-impaired patients.

[0003] Automatic gain control (AGaS) in hearing aids can use linear amplification for soft and medium sounds, and compression amplification for sounds above 65dB (medium) at high volumes. However, in real-world applications, the acquired speech signal often contains ambient noise, affecting the accuracy of speech transmission. The effectiveness of AGaS in hearing aids depends on the noise reduction of the sound signal. Separating the target sound source signal from the noise signal significantly improves AGaS performance. If a hearing aid wearer is interested in someone's speech, they will primarily hear that person's voice, ignoring other surrounding sounds. This process of extracting unknown source signals from a large number of mixed signals is called blind source separation (BSS), which provides technical support for speech signal enhancement and improves the sound quality of hearing aids.

[0004] The natural gradient algorithm is a commonly used method in blind source separation, characterized by fast convergence and good separation performance. However, the step size of the natural gradient algorithm is fixed. If the step size is too large, the steady-state error will be large; if the step size is too small, the convergence speed will be too slow. In the early stages of sound signal separation, the correlation between sound signals is generally high, so a larger step size can be used. As the separation degree increases, the correlation between sound signals decreases, and a smaller step size should be used. Therefore, using a natural gradient algorithm with a fixed step size will result in poor separation of sound source signals and ambient noise signals, and will not meet the real-time requirements of hearing aids, affecting the user experience of hearing-impaired individuals. Summary of the Invention

[0005] To address the aforementioned technical problems, the purpose of this application is to provide a method and system for automatic gain control of a digital hearing aid, the specific technical solution of which is as follows: In a first aspect, embodiments of this application provide an automatic gain control method for a digital hearing aid, the method comprising the following steps: The sound signal is collected through the built-in microphone of the hearing aid; the Melanographic spectrogram of the sound signal is obtained. Each frame of the Mel spectrogram is divided into Mel frequency bands; the energy concentration of each Mel frequency band is determined by the energy value and local density of each coordinate point in each Mel frequency band, combined with the metric distance between adjacent coordinate points in each Mel frequency band. The differences between high-frequency and low-frequency energies in each frame of the Mel speech spectrogram, as well as the energy distribution in each frame of the Mel speech spectrogram, are analyzed. Combined with the energy concentration, the noise significance of each frame of the Mel speech spectrogram is determined. The natural gradient algorithm is used to perform blind source separation on the acquired sound signal. The separation difference degree of each iteration is obtained by combining the similarity between the target signal and the noise signal obtained after each iteration, the similarity between the acquired sound signal and the separated signal, and the noise saliency of all frames of Mel spectrograms, so as to determine the step size of the next iteration. Automatic gain control is applied to the target signal obtained from blind source separation.

[0006] In one embodiment, determining the energy concentration of each Mel frequency band includes: Calculate the product of the energy value and its local density at each coordinate point in each Mel frequency band, determine the fusion result of the product of all coordinate points in each Mel frequency band, and record it as the first fusion value; determine the fusion result of the metric distance between all adjacent coordinate points in each Mel frequency band, and record it as the second fusion value. The energy concentration is determined based on the first fusion value and the second fusion value.

[0007] In one embodiment, the energy concentration is positively correlated with the first fusion value and negatively correlated with the second fusion value.

[0008] In one embodiment, determining the noise saliency of each frame of Mel spectrogram includes: Divide all Mel frequency bands in each frame of Mel spectrogram into low frequency and high frequency parts, and calculate the ratio of the sum of energy and value of all coordinate points in all Mel frequency bands in the high frequency range to the sum of energy and value of all coordinate points in all Mel frequency bands in the low frequency range. Determine the average energy of all coordinate points in each frame of the Mel language spectrogram; The fusion result of the energy concentration of all Mel frequency bands in each frame of Mel spectrogram is determined and denoted as the third fusion value; The noise significance is positively correlated with the ratio, the mean energy, and the third fusion value.

[0009] In one embodiment, the noise significance is the normalized result of the product of the ratio, the mean energy, and the third fusion value.

[0010] In one embodiment, obtaining the separation difference degree for each iteration includes: Determine the proportion of the number of iterations corresponding to each iteration of the natural gradient algorithm in the preset total number of iterations, and calculate the average noise significance of all frames of Mel spectrograms; The similarity between the target signal and the noise signal obtained after each iteration is denoted as the first similarity. The fusion result of the similarity between the collected sound signal and all the separated signals is denoted as the fourth fusion value. The separation difference is determined by the proportion, the average value, and the first similarity and the fourth fusion value.

[0011] In one embodiment, the inverse proportional mapping result between the first similarity and the fourth fusion value is obtained, and then positively fused with the proportion and the average value to obtain the separation difference.

[0012] In one embodiment, determining the step size for the next iteration includes: The normalized result of the separation difference in each iteration is multiplied by the preset step size adjustment sensitivity, and combined with the step size of each iteration, the step size of the next iteration is obtained.

[0013] In one embodiment, the difference between the natural number 1 and the result of the multiplication is calculated, and the step size of the next iteration is the product of the step size of each iteration and the difference.

[0014] Secondly, embodiments of this application also provide an automatic gain control system for a digital hearing aid, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the steps of any of the methods described above.

[0015] This application has at least the following beneficial effects: This application acquires sound signals via a built-in microphone in a hearing aid; obtains a Mel spectrogram of the sound signal; divides each frame of the Mel spectrogram into Mel frequency bands, determines the energy concentration of each Mel frequency band, and accurately quantifies the energy distribution in each Mel frequency band, which helps to analyze the noise content in each frame of the Mel spectrogram and improves the reliability of subsequent noise saliency calculation; furthermore, it determines the noise saliency of each frame of the Mel spectrogram, reflecting the noise content in each frame of the Mel spectrogram and improving the accuracy of noise extraction in each frame of the Mel spectrogram; and uses the similarity between the target signal and the noise signal obtained after each iteration of the natural gradient algorithm, as well as the similarity between the acquired sound signal and the separated sound signal, to determine the noise content in each frame of the Mel spectrogram. The similarity of the signals, combined with the noise saliency of all frames of Mel spectrograms, yields the separation difference for each iteration, which determines the step size for the next iteration. This ensures that the step size of each iteration incorporates the signal separation effect of the previous iteration, improving the accuracy and suitability of step size determination in the natural gradient algorithm. This further enhances the accuracy of blind source separation, avoids signal distortion, optimizes the automatic gain control effect of the hearing aid, improves the clarity of the target signal, and enhances the applicability and comfort of the hearing aid. It helps provide users with a more natural auditory experience, effectively improves the performance of the hearing aid in noisy environments, and solves the problem of users not being able to clearly hear conversations or surrounding sounds in complex environments. Attached Figure Description

[0016] 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.

[0017] Figure 1 A flowchart illustrating the steps of an automatic gain control method for a digital hearing aid, as provided in one embodiment of this application; Figure 2 A flowchart for determining the step size in the natural gradient algorithm iteration. Detailed Implementation

[0018] To further illustrate the technical means and effects adopted by this application to achieve the intended inventive purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a digital hearing aid automatic gain control 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.

[0019] 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.

[0020] The following description, in conjunction with the accompanying drawings, details the specific scheme of the automatic gain control method and system for a digital hearing aid provided in this application.

[0021] Please see Figure 1 The diagram illustrates a flowchart of an automatic gain control method for a digital hearing aid according to an embodiment of this application. The method includes the following steps: S1 collects sound signals through the built-in microphone of the hearing aid; and obtains the Mel-language spectrogram of the sound signal.

[0022] Hearing aid devices mainly consist of microphones, analog-to-digital converters (ADCs), digital processing chips, and speaker structures. When a hearing aid wearer converses with someone, the hearing aid first collects the received sound signal through its two built-in microphones. This sound signal includes the other person's voice and ambient noise signals; therefore, the sound signal collected by the microphones is referred to as a mixed signal. Its sampling frequency is 32kHz, which can be set by the user according to the actual situation. Then, the mixed signal is converted into a digital sound signal through an analog-to-digital converter (ADC). Since the actual conversation environment is complex and diverse, with low-frequency noise (such as wind noise) and high-frequency noise (such as electronic interference), the digital signal, when passing through the digital processing chip, uses an embedded Butterworth filter to initially remove the noise effects, outputting a digital sound signal containing both the sound source signal and ambient noise signals, referred to as a mixed digital signal. Subsequently, this mixed digital signal is separated to separate the target signal that the wearer wants to hear, i.e., the sound source signal, and the gain of the target signal is automatically controlled to improve the sound quality. Among them, the AD analog-to-digital conversion and Butterworth filter are both existing well-known technologies, and the specific process will not be described in detail.

[0023] When hearing aid wearers talk to people outdoors, there are often surrounding noises, such as car horns and wind, causing the sound signal collected by the hearing aid to contain a large amount of noise, which can be simply referred to as... ,in, The observed signal is the mixed digital signal captured by the microphone, and A is the mixing matrix, representing the mixing characteristics of the signal. The source signal, i.e., the actual multi-source sound signal in the mixed digital signal, is used in this embodiment. These are the actual target signal and noise signal. Blind source separation is based solely on the observed signal. Find the separation matrix W, where W is the inverse of A, such that the transformed output... It is the source signal An estimate.

[0024] Since the target signal and noise signal are independent of each other, the energy of the sound signal generated by the same person speaking will usually be concentrated in the same frequency range on the vertical axis of the Mel speech spectrogram, and this frequency range changes smoothly over time. The energy distribution of adjacent time frames has a strong correlation, and the color changes smoothly. However, the noise signal that suddenly appears in the surrounding area will have higher energy, and the energy distribution on the vertical axis of the Mel speech spectrogram will be wider, significantly exceeding the high-frequency area covered by the normal speaking sound signal. When the noise signal appears at a certain moment, the Mel speech spectrogram will show a bright area at that moment, that is, a high energy, which usually lasts for a certain period of time on the horizontal axis, forming a short-term energy distribution segment. Blind source separation is required for the mixed digital signal during this period.

[0025] Based on the above analysis, the Mel spectrogram of the mixed digital signal is obtained. The mixed digital signal is processed by frame-by-frame windowing with a frame length of 10ms, a step size of 5ms, and a Hamming window function. Short-time Fourier transform is used for spectrum calculation, and RGB values ​​are used for amplitude quantization. The Mel-language spectrogram is converted to a known technique; specific implementation details are not elaborated here. In the Mel-language spectrogram, the horizontal axis represents time (frame number), and the vertical axis represents frequency. The coordinates of the i-th row and j-th column of the Mel-language spectrogram are... The amplitude corresponding to this coordinate in the Melan spectrum is denoted as the energy value. , representing the signal energy at frequency j in the mixed digital signal of the i-th frame, i.e., the sound energy value, is represented by the Mel spectrogram. The i-th frame is denoted as In obtaining the Mel language spectrogram The implementer can set the frame length and step size according to the actual situation, and this embodiment does not impose any restrictions on this.

[0026] S2, divide each frame of Mel spectrogram into Mel frequency bands; determine the energy concentration of each Mel frequency band by the energy value and local density of each coordinate point in each Mel frequency band, combined with the metric distance between adjacent coordinate points in each Mel frequency band.

[0027] When two people are talking normally, the fluctuations in their voices are relatively small, and the corresponding sound signal frequency and energy will also fluctuate within a certain range. However, sudden sharp noise usually has a large sound energy value, and the sound signal shows obvious peak fluctuations. It has more high-frequency energy in the Mel spectrogram, and the energy density at each coordinate point is greater. Based on this, the distribution area of ​​sharp noise in the Mel spectrogram can be determined.

[0028] Based on the above analysis, this embodiment uses the density peak clustering algorithm to calculate the coordinates of each coordinate point in the Melanographic spectrogram of each frame. The local density corresponding to the energy value is denoted as the local energy density. The difference in energy values ​​between each coordinate point is used as the distance between coordinate points in the density peak clustering algorithm, a well-known existing technique whose specific process will not be elaborated upon. Then, the Mel spectrogram corresponding to each frame is... The frequency bands are evenly divided into N Mel frequency bands, and the Mel frequency bands are numbered sequentially from low frequency to high frequency. Each frequency band is denoted as . , In this embodiment, N=20. Implementers can set it according to the actual situation. This embodiment does not impose any restrictions on this.

[0029] Based on this, the energy concentration of each Mel frequency band in each frame of the Mel spectrogram is calculated to characterize the energy distribution in that Mel frequency band, specifically: Calculate the product of the energy value and its local density at each coordinate point in each Mel frequency band, determine the fusion result of the product of all coordinate points in each Mel frequency band, and record it as the first fusion value; determine the fusion result of the metric distance between all adjacent coordinate points in each Mel frequency band, and record it as the second fusion value. The energy concentration is determined based on the first fusion value and the second fusion value, wherein the energy concentration is positively correlated with the first fusion value and negatively correlated with the second fusion value.

[0030] It should be noted that fusion means combining multiple variables, which can be done by addition, multiplication, a combination of addition and multiplication, or by taking the average.

[0031] In this embodiment, the expression for the energy concentration of each Mel frequency band in each frame of the Mel spectrogram is: In the formula, This represents the energy concentration of the nth Mel frequency band in the Mel spectrogram of the i-th frame. This indicates the number of coordinate points contained in each Mel frequency band. This represents the k-th coordinate point. Let k represent the local energy density at the k-th coordinate point in the n-th Mel frequency band of the i-th frame Mel spectrogram. This represents the energy value corresponding to the k-th coordinate point in the n-th Mel frequency band of the i-th frame Mel spectrogram. , These represent the k-th and (k+1)-th coordinates in the n-th Mel frequency band of the i-th frame Mel spectrogram, respectively. This indicates that the distance measurement is performed. In this embodiment, Euclidean distance is used for calculation. However, implementers can choose other existing feasible distance measurement methods, and this embodiment does not impose any restrictions on this. This is recorded as the first fusion value. This is denoted as the second fusion value.

[0032] S3. Analyze the difference between high-frequency and low-frequency energy in each frame of the Mel speech spectrogram, as well as the energy distribution in each frame of the Mel speech spectrogram. Combined with the energy concentration, determine the noise significance of each frame of the Mel speech spectrogram.

[0033] Furthermore, the noise saliency of each frame of the Mel spectrogram is calculated to characterize the probability that each frame of the Mel spectrogram contains sharp noise, specifically: Each frame of the Mel spectrogram is divided into two parts: low frequency and high frequency. In this embodiment, each frame of the Mel spectrogram is divided into N Mel frequency bands, and these N Mel frequency bands are arranged in order from low frequency to high frequency. Therefore, the first N / 2 Mel frequency bands are marked as low frequency bands, and the last N / 2 Mel frequency bands are marked as high frequency bands.

[0034] In this embodiment, the expression for the noise saliency of each frame of Mel spectrogram is: In the formula, This represents the noise saliency of the Melanographic spectrogram in the i-th frame. Represents the normalization function. This represents the ratio of the sum of energy and values ​​of all coordinate points in all high-frequency bands to the sum of energy and values ​​of all coordinate points in all low-frequency bands in the Melan language spectrogram of frame i. Let be the average energy of all coordinate points in the Melan language spectrogram of the i-th frame. This represents the energy concentration of the nth Mel frequency band in the Mel spectrogram of the i-th frame, where N is the number of all Mel frequency bands in each frame of the Mel spectrogram; in this embodiment, N=20. This is denoted as the third fusion value.

[0035] The greater the energy concentration in each Mel frequency band of the i-th frame Mel spectrogram, the more high-frequency energy there is, and the greater the corresponding noise significance, indicating that the i-th frame Mel spectrogram is more likely to contain sharp noise.

[0036] S4 uses the natural gradient algorithm to perform blind source separation on the acquired sound signal. By combining the similarity between the target signal and the noise signal obtained after each iteration, and the similarity between the acquired sound signal and the separated signal, along with the noise saliency of all frames of Mel spectrograms, the separation difference of each iteration is obtained to determine the step size of the next iteration.

[0037] When using the natural gradient algorithm to perform blind source separation of digital mixed signals, if the overall noise significance of the digital mixed signal is greater, it means that the degree of separation of the digital mixed signal is greater with the iteration of the natural gradient algorithm, and the difference between the separated target signal and the noise signal is greater after each iteration. Therefore, the iteration step size should be adaptively updated and reduced to improve the separation accuracy and avoid the loss of the target signal due to the excessive step size, which would cause the separated target signal to be distorted.

[0038] Based on the above analysis, the separation difference between the target signal and the noise signal separated by the natural gradient algorithm after each iteration is calculated. This is used to characterize the degree of separation and difference between the two signals separated from the digital mixed signal. Specifically: Determine the proportion of the number of iterations corresponding to each iteration of the natural gradient algorithm in the preset total number of iterations, and calculate the average noise significance of all frames of Mel spectrograms; The similarity between the target signal and the noise signal obtained after each iteration is denoted as the first similarity. The fusion result of the similarity between the collected sound signal and all the separated signals is denoted as the fourth fusion value. Obtain the inverse proportional mapping result between the first similarity and the fourth fusion value, and perform positive fusion with the proportion and the average value to obtain the separation difference.

[0039] It should be noted that the similarity between signals can be calculated using methods such as Pearson correlation coefficient and cosine similarity; inverse proportional mapping indicates a mapping method in which there is an inverse proportional relationship between the dependent variable and the independent variable, that is, the dependent variable will decrease as the independent variable increases, and will also increase as the independent variable decreases, such as reciprocal relationship, negative exponential relationship, etc.

[0040] In this embodiment, the expression for the separation difference is: In the formula, This represents the degree of separation difference between the target signal and the noise signal after the d-th iteration of the natural gradient algorithm. This indicates the d-th iteration. This represents the total number of iterations. In this embodiment, D=100, but the implementer can set it according to the actual situation. This indicates the total number of frames in the Mel language spectrogram. , Let them represent the target signal and the noise signal separated after the d-th iteration of the natural gradient algorithm, respectively. This represents the total number of signal types separated by the natural gradient algorithm. In this example, Q=2, indicating that two types of signals are separated: the target signal and the noise signal. Indicates a mixed digital signal. This indicates the calculation of the Pearson correlation coefficient. This represents the q-th signal separated after the d-th iteration of the natural gradient algorithm. This represents the noise saliency of the Melanographic spectrogram in the i-th frame. This is denoted as the first similarity. This is denoted as the second similarity. This is denoted as the fourth fusion value.

[0041] It should be understood that, A positive correlation exists between noise significance and the magnitude of noise in the Mel spectrogram. The more significant the noise, the greater the difference between the separated target signal and the noise signal when the natural gradient algorithm separates mixed digital signals; that is, the greater the separation difference. Furthermore, as the iteration progresses... As the value increases, the correlation coefficient between the target signal and the noise signal decreases. And the correlation coefficient between the mixed digital signal and the separated signal. It will gradually decrease, then As the value increases, it indicates that the degree of signal separation is increasing.

[0042] Furthermore, based on the separation difference between the target signal and the noise signal after each iteration, the step size of the natural gradient algorithm in the next iteration is adaptively adjusted, specifically as follows: In the formula, This represents the step size in the d-th iteration of the natural gradient algorithm. This represents the step size in the (d-1)th iteration of the natural gradient algorithm. The preset step size adjustment sensitivity has a value range of 0-1. The larger the value, the higher the sensitivity of the step size adjustment. Relative to the step size of the (d-1)th iteration, the larger the change in step size in the d-th iteration. In this embodiment... The implementer can set it according to the actual situation, and this embodiment does not impose any restrictions on it. Norm() represents the separation difference between the target signal and the noise signal after the (d-1)th iteration of the natural gradient algorithm, and Norm() is the normalization function. In this embodiment, the initial step size of the natural gradient algorithm iteration is 0.05, but the implementer can set it according to the actual situation. The flowchart for determining the step size of the natural gradient algorithm iteration is shown below. Figure 2 As shown.

[0043] It should be understood that the greater the difference between the target signal and the noise signal separated by the natural gradient algorithm after each iteration, the smaller the iteration step size should be in the next iteration to improve the separation accuracy and avoid the loss of the target signal due to the excessive step size, which would cause the separated target signal to be distorted.

[0044] A natural gradient algorithm with adaptive step size update is used to perform blind source separation on mixed digital signals. Based on the results of blind source separation, speech activity detection (VAD) technology is used to identify the target signal in the mixed digital signals. Both the natural gradient algorithm for blind source separation and VAD are existing well-known technologies, and their specific processes are not detailed here.

[0045] S5 performs automatic gain control on the target signal obtained from blind source separation.

[0046] The target signal obtained from blind source separation is first restored to its original sound pressure level range through amplitude correction to obtain the original speech signal. Then, according to the hearing aid's automatic gain control requirements, a gain is applied to the original speech signal. For example, linear amplification is performed on the weak sound band with a gain limit of 20 dB, and nonlinear compression is performed on the strong sound band with a compression ratio of 1:2 to ensure that the sound output from the hearing aid is delivered clearly and comfortably to the hearing-impaired person's ear. Amplitude correction and automatic gain control are both existing well-known technologies, and their specific implementation details will not be elaborated here.

[0047] Based on the same inventive concept as the above methods, this application also provides a digital hearing aid automatic gain control 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 automatic gain control methods.

[0048] 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.

[0049] 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.

[0050] 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 automatic gain control in a digital hearing aid, characterized in that, The method includes the following steps: The sound signal is collected through the built-in microphone of the hearing aid; the Melanographic spectrogram of the sound signal is obtained. Each frame of the Mel spectrogram is divided into Mel frequency bands; the energy concentration of each Mel frequency band is determined by the energy value and local density of each coordinate point in each Mel frequency band, combined with the metric distance between adjacent coordinate points in each Mel frequency band. The differences between high-frequency and low-frequency energies in each frame of the Mel speech spectrogram, as well as the energy distribution in each frame of the Mel speech spectrogram, are analyzed. Combined with the energy concentration, the noise significance of each frame of the Mel speech spectrogram is determined. The natural gradient algorithm is used to perform blind source separation on the acquired sound signal. The separation difference degree of each iteration is obtained by combining the similarity between the target signal and the noise signal obtained after each iteration, the similarity between the acquired sound signal and the separated signal, and the noise saliency of all frames of Mel spectrograms, so as to determine the step size of the next iteration. Automatic gain control is applied to the target signal obtained from blind source separation.

2. The automatic gain control method for a digital hearing aid as described in claim 1, characterized in that, Determining the energy concentration of each Mel frequency band includes: Calculate the product of the energy value and its local density at each coordinate point in each Mel frequency band, determine the fusion result of the product of all coordinate points in each Mel frequency band, and record it as the first fusion value; determine the fusion result of the metric distance between all adjacent coordinate points in each Mel frequency band, and record it as the second fusion value. The energy concentration is determined based on the first fusion value and the second fusion value.

3. The automatic gain control method for a digital hearing aid as described in claim 2, characterized in that, The energy concentration is positively correlated with the first fusion value and negatively correlated with the second fusion value.

4. The automatic gain control method for a digital hearing aid as described in claim 1, characterized in that, Determining the noise saliency of each frame of Mel spectrogram includes: Divide all Mel frequency bands in each frame of Mel spectrogram into low frequency and high frequency parts, and calculate the ratio of the sum of energy and value of all coordinate points in all Mel frequency bands in the high frequency range to the sum of energy and value of all coordinate points in all Mel frequency bands in the low frequency range. Determine the average energy of all coordinate points in each frame of the Mel language spectrogram; The fusion result of the energy concentration of all Mel frequency bands in each frame of Mel spectrogram is determined and denoted as the third fusion value; The noise significance is positively correlated with the ratio, the mean energy, and the third fusion value.

5. The automatic gain control method for a digital hearing aid as described in claim 4, characterized in that, The noise significance is the normalized result of the product of the ratio, the mean energy, and the third fusion value.

6. The automatic gain control method for a digital hearing aid as described in claim 1, characterized in that, The process of obtaining the separation difference degree for each iteration includes: Determine the proportion of the number of iterations corresponding to each iteration of the natural gradient algorithm in the preset total number of iterations, and calculate the average noise significance of all frames of Mel spectrograms; The similarity between the target signal and the noise signal obtained after each iteration is denoted as the first similarity. The fusion result of the similarity between the collected sound signal and all the separated signals is denoted as the fourth fusion value. The separation difference is determined by the proportion, the average value, and the first similarity and the fourth fusion value.

7. The automatic gain control method for a digital hearing aid as described in claim 6, characterized in that, Obtain the inverse proportional mapping result between the first similarity and the fourth fusion value, and perform positive fusion with the proportion and the average value to obtain the separation difference.

8. The automatic gain control method for a digital hearing aid as described in claim 1, characterized in that, The step size for determining the next iteration includes: The normalized result of the separation difference in each iteration is multiplied by the preset step size adjustment sensitivity, and combined with the step size of each iteration, the step size of the next iteration is obtained.

9. The automatic gain control method for a digital hearing aid as described in claim 8, characterized in that, Calculate the difference between the natural number 1 and the result of multiplying them. The step size of the next iteration is the product of the step size of each iteration and the difference.

10. A digital hearing aid automatic gain control 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 method as described in any one of claims 1-9.

Images