A digital hearing aid self-sound detection adjusting method based on air-bone fusion

Abstract

The application discloses a digital hearing aid self-sound detection adjusting method based on air-bone conduction fusion, which comprises the following steps: signal acquisition and pretreatment; air conduction signal strength and frequency domain calculation; bone conduction signal strength and frequency domain calculation; gain generation and signal output. By introducing a bone conduction vibration sensor, the signal amplitude of the wearer's own voice is reduced when the wearer speaks, and the energy is compressed. Meanwhile, when the input signal is only the voice of others in the environment, the signal amplitude will not be reduced, and the voice of others will not be inhibited, so that the energy of the wearer's own voice is adjusted to a reasonable level, the hearing feeling of the wearer when speaking is improved, and compared with the traditional wide dynamic range compression algorithm in the digital hearing aid, the self-sound detection and inhibition effect is better.

Description

Technical Field

[0001] This invention belongs to the field of signal processing technology, specifically relating to a method for self-sound detection and adjustment of digital hearing aids based on air-bone conduction fusion. Background Technology

[0002] In recent years, with the increasing severity of aging in China, the number of people with hearing impairments has been rising. Hearing aids can improve the hearing conditions of people with hearing impairments, enabling patients of different types and ages to quickly receive personalized compensation for different degrees of hearing loss, thereby hearing external sounds better.

[0003] Currently, common hearing aids are mainly divided into digital hearing aids and analog hearing aids. Compared with analog hearing aids, digital hearing aids have the following obvious advantages: digital hearing aids can achieve more refined time-domain and frequency-domain adjustment, support more complex and advanced algorithms such as feedback suppression and speech enhancement, and support greater degree of customization. Digital hearing aids and related technologies have broad application and development prospects in the field of hearing health. Therefore, research on programmable digital hearing aid algorithms is of profound significance.

[0004] However, when a user wears a hearing aid, their own voice is picked up by the hearing aid's microphone and amplified without discrimination before being transmitted into the user's ear through the speaker. Because this part of the sound has a large energy and does not need to be amplified, it ultimately leads to a serious decline in the wearing experience of the hearing aid. Therefore, how to improve the problem of the amplification of one's own voice during the wearing of digital hearing aids, which causes a decline in the listening experience, is an urgent problem to be solved. To this end, we propose a self-sound detection and adjustment method for digital hearing aids based on air-bone conduction fusion. Summary of the Invention

[0005] The purpose of this invention is to provide a self-sound detection and adjustment method for digital hearing aids based on air-bone conduction fusion, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a self-sound detection and adjustment method for a digital hearing aid based on air-bone conduction fusion, comprising the following steps: Step 1, signal acquisition and preprocessing: acquiring continuous air conduction signals through a first vibration sensor and continuous bone conduction signals through a second vibration sensor, respectively; performing frame-by-frame filtering preprocessing on the two continuous signals to obtain preprocessed framed air conduction time-domain signals and framed bone conduction time-domain signals; Step 2, air conduction signal strength and frequency domain calculation: sending the preprocessed framed air conduction time-domain signals to an air conduction signal strength detection unit to calculate the air conduction signal probability, and simultaneously performing frequency domain calculation on the signal. Step 3: Bone conduction signal strength and frequency domain calculation: The preprocessed framed bone conduction time domain signal is sent to the bone conduction signal strength detection unit to calculate the bone conduction signal probability. At the same time, a frequency domain transformation is performed on the signal, and the frequency domain energy distribution sequence of the bone conduction signal is obtained through the frequency domain detection unit. Step 4: Gain generation and signal output: The probability of the air conduction signal and the probability of the bone conduction signal are combined to generate a frequency domain gain sequence. The frequency domain gain sequence is then subjected to clipping, limiting and smoothing processing to obtain the application gain. The application gain is applied to the air conduction frequency domain information and then subjected to inverse frequency domain transformation to output the enhanced time domain air conduction signal.

[0007] Preferably, in step 1, the air conduction continuous signal acquired by the first vibration sensor is framed to obtain a framed air conduction signal. The framed air conduction signal is input into the first air conduction signal filter to obtain a framed air conduction time domain signal with DC offset component removed. The bone conduction continuous signal acquired by the second vibration sensor is framed to obtain a framed bone conduction signal. The framed bone conduction signal is first input into the first bone conduction signal filter to obtain a framed bone conduction signal with DC offset component removed for the first time, and then input into the second bone conduction signal filter to obtain a framed bone conduction time domain signal with high frequency component removed.

[0008] Preferably, in step 2, the specific process for calculating the probability of the air conduction signal is as follows: Step 2.1, calculate the current frame power of the framed air conduction time domain signal, as shown in formula (1): (1);

[0009] in, The m-th frame represents the air-conduction time-domain signal power; n is the sample number; N is the fixed frame length. This represents the square of the nth sample value of the filtered air conduction signal; Step 2.2: Calculate the logarithmic weighted average power of historical frames, as shown in formula (2): (2);

[0010] in, denoted as the log-weighted average power of the historical frames of the air conduction signal; K represents the number of historical frames selected; m represents the current frame position; Step 2.3: Input the squared value of the log-weighted average power of the historical frames of the segmented air conduction time domain signal and the log-weighted average power of the historical frames into a first-order infinite impulse response filter to obtain the smoothed power value, as shown in formulas (3) and (4): (3);

[0011] (4);

[0012] in, This represents the log-weighted average squared power value of the historical frames of the smoothed air conduction signal; This represents the log-weighted average power of the smoothed air conduction signal across historical frames. This represents the smoothing coefficient used to calculate the logarithmically weighted average power of the historical frames of the air conduction signal. Values: 63 / 64; This represents the smoothing coefficient used to calculate the log-weighted average squared power value of the historical frames of the air conduction signal. Take a value of 3 / 4; Step 2.4: Update the air conduction signal probability frame by frame based on the smoothed power value, as shown in formula (5): (5);

[0013] This represents the smoothing coefficient used to calculate the probability of the air conduction signal. The value is 7 / 8.

[0014] Preferably, in step 2, the frequency domain information of the air conduction signal is obtained by performing a frequency domain transformation on the preprocessed framed air conduction time domain signal, and the frequency domain power spectrum sequence of the air conduction signal is calculated by combining the frequency domain transformation result, as shown in formula (6): (6);

[0015] This represents the frequency domain power spectrum sequence of the air conduction signal, where m is the current frame number and j is the frequency point number; This represents the complex spectrum of the filtered air conduction signal at the j-th frequency point in the m-th frame after frequency domain transformation.

[0016] Preferably, in step 3, the specific process for calculating the bone conduction signal probability is as follows: based on the framed bone conduction time-domain signal after secondary filtering, the current frame power, the logarithmic weighted average power of historical frames, and the smoothed power value are calculated sequentially, and the bone conduction signal probability is updated frame by frame; wherein, the current frame power of the framed bone conduction time-domain signal is calculated as shown in formula (7): (7);

[0017] in, This represents the bone conduction time-domain signal power in the m-th frame; The value represents the square of the nth sample point of the filtered bone conduction signal; the log-weighted average power of historical frames is calculated as shown in formula (8): (8);

[0018] in, This represents the log-weighted average power of the bone conduction signal across historical frames. Let represent the bone conduction time-domain signal power of the nth frame; input the squared value of the logarithmic weighted average power of the historical frames of the framed bone conduction time-domain signal and the logarithmic weighted average power of the historical frames into a first-order infinite impulse response filter to obtain the smoothed power value, as shown in formulas (9) and (10): (9);

[0019] (10);

[0020] in, This represents the log-weighted average power of the smoothed bone conduction signal across historical frames. This represents the log-weighted average power squared value of the bone conduction signal from historical frames after smoothing. This represents the smoothing coefficient used to calculate the log-weighted average power of the bone conduction signal across historical frames. Values: 63 / 64; This represents the smoothing coefficient used to calculate the log-weighted average squared power value of the bone conduction signal across historical frames. The value is 3 / 4; the bone conduction signal probability is obtained by updating the smooth power value frame by frame, as shown in formula (11): (11);

[0021] in, This represents the smoothing coefficient used to calculate the probability of bone conduction signals. The value is 7 / 8.

[0022] Preferably, in step 3, the frequency domain information of the bone conduction signal is obtained by performing a frequency domain transformation on the preprocessed framed bone conduction time domain signal, and the frequency domain power spectrum sequence of the bone conduction signal is calculated by combining the frequency domain transformation result, as shown in formula (12): (12);

[0023] in, This represents the frequency domain power spectrum sequence of the bone conduction signal; This represents the complex spectrum of the filtered bone conduction signal at the j-th frequency point in the m-th frame after frequency domain transformation.

[0024] Preferably, in step 4, the air conduction signal probability and the bone conduction signal probability are combined to generate a frequency domain gain sequence. The frequency domain gain sequence is then subjected to clipping, limiting, and smoothing processes to obtain the application gain. After the application gain is applied to the air conduction frequency domain information, the enhanced time-domain air conduction signal is output through inverse frequency domain transformation. The specific steps are as follows: Step 4.1: Generate the frequency domain gain as shown in formula (13): (13);

[0025] Where D represents the gain adjustment constant; Step 4.2, the frequency domain gain sequence is clipped and limited, as shown in formula (14): (14);

[0026] Step 4.3: Smooth the clipped and limited gain sequence as shown in formula (15): (15);

[0027] Step 4.4: Multiply the applied gain with the air conduction frequency domain information, perform an inverse frequency domain transform on the multiplication result, and finally output the enhanced time-domain air conduction signal after suppressing self-acoustic sound.

[0028] Compared with the prior art, the beneficial effects of the present invention are as follows: Based on the air conduction microphone, the present invention introduces a bone conduction vibration sensor to reduce the signal amplitude and compress the energy of the wearer's own voice when the wearer speaks. At the same time, when the input signal only contains the voices of others in the environment, the signal amplitude will not decrease and the voices of others will not be suppressed. This achieves the adjustment of the energy of the wearer's own voice to a reasonable level, improving the wearer's hearing experience when speaking. Compared with the traditional wide dynamic range compression algorithm in digital hearing aids, it can provide better self-sound detection and suppression effects. Attached Figure Description

[0029] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0030] Figure 2 This is a time-domain waveform diagram of the input signal for the self-sound detection and adjustment method of the present invention;

[0031] Figure 3 This is a probability diagram of the wearer's own voice output by the self-voice detection and adjustment method of the present invention.

[0032] Figure 4 This is a time-domain waveform diagram of the input signal for the self-sound detection and adjustment method of the present invention;

[0033] Figure 5 This is a time-domain waveform diagram of the output signal after self-sound adjustment according to the present invention;

[0034] Figure 6 This is the spectrogram of the input signal for the self-sound detection and adjustment method of the present invention;

[0035] Figure 7 This is the spectrogram of the output signal after self-sound adjustment according to the present invention;

[0036] Figure 8 The time-domain waveform of the input signal that speaks itself;

[0037] Figure 9 The time-domain waveform of the output signal after adjusting the self-voice (which has its own voice);

[0038] Figure 10 The spectrogram of the input signal that exists while speaking;

[0039] Figure 11 The spectrogram of the output signal after the self-voice has been adjusted to reflect the speaker's own speech;

[0040] Figure 12 The time-domain waveform of the input signal where only others are speaking;

[0041] Figure 13 The time-domain waveform of the output signal when only others are speaking;

[0042] Figure 14 The spectrogram of the input signal where only others are speaking;

[0043] Figure 15 The spectrogram of the output signal after adjusting the self-voice when only others are speaking. Detailed Implementation

[0044] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0045] Please see Figure 1The present invention provides a self-sound detection and adjustment method for digital hearing aids based on air-bone conduction fusion, comprising the following steps: Step 1, signal acquisition and preprocessing: air conduction continuous signal is acquired by a first vibration sensor and bone conduction continuous signal is acquired by a second vibration sensor. The two continuous signals are framed and then filtered and preprocessed to obtain preprocessed framed air conduction time-domain signal and framed bone conduction time-domain signal. Compared with directly using the acquired signal, filtering and preprocessing the two continuous signals separately is to facilitate the extraction of signal features in subsequent steps. In Step 1, the air conduction continuous signal acquired by the first vibration sensor is framed to obtain a framed air conduction signal. The framed air conduction signal is input into the first air conduction signal filter to obtain a framed air conduction time-domain signal with DC offset component removed. The bone conduction continuous signal acquired by the second vibration sensor is framed to obtain a framed bone conduction signal. The framed bone conduction signal is first input into the first bone conduction signal filter to obtain a framed bone conduction signal with DC offset component removed for the first time, and then input into the second bone conduction signal filter to obtain a framed bone conduction time-domain signal with high frequency component removed.

[0046] Step 2, Calculation of air conduction signal strength and frequency domain: The preprocessed framed air conduction time domain signal is sent to the air conduction signal strength detection unit to calculate the air conduction signal probability. At the same time, the frequency domain transformation is performed on the signal to obtain the air conduction signal frequency domain information. Compared with analyzing only the single time domain information, the frequency domain transformation of the signal can obtain more frequency domain feature information, which is convenient for analyzing the energy part of the signal in different frequency bands. In step 2, the specific process of calculating the air conduction signal probability is as follows: Step 2.1, Calculate the current frame power of the framed air conduction time domain signal, as shown in formula (1): (1);

[0047] in, The m-th frame represents the air-conduction time-domain signal power; n is the sample number; N is the fixed frame length. This represents the square of the nth sample value of the filtered air conduction signal; Step 2.2: Calculate the logarithmic weighted average power of historical frames, as shown in formula (2): (2);

[0048] in, denoted as the log-weighted average power of the historical frames of the air conduction signal; K represents the number of historical frames selected; m represents the current frame position; Step 2.3: Input the squared value of the log-weighted average power of the historical frames of the segmented air conduction time domain signal and the log-weighted average power of the historical frames into a first-order infinite impulse response filter to obtain the smoothed power value, as shown in formulas (3) and (4): (3);

[0049] (4);

[0050] in, This represents the log-weighted average squared power value of the historical frames of the smoothed air conduction signal; This represents the log-weighted average power of the smoothed air conduction signal across historical frames. This represents the smoothing coefficient used to calculate the logarithmically weighted average power of the historical frames of the air conduction signal. Values: 63 / 64; This represents the smoothing coefficient used to calculate the log-weighted average squared power value of the historical frames of the air conduction signal. Take a value of 3 / 4; Step 2.4: Update the air conduction signal probability frame by frame based on the smoothed power value, as shown in formula (5): (5);

[0051] This represents the smoothing coefficient used to calculate the probability of the air conduction signal. The value is 7 / 8;

[0052] In step 2, the frequency domain information of the air conduction signal is obtained by performing a frequency domain transformation on the preprocessed framed air conduction time domain signal, and the frequency domain power spectrum sequence of the air conduction signal is calculated by combining the frequency domain transformation result, as shown in formula (6): (6);

[0053] This represents the frequency domain power spectrum sequence of the air conduction signal, where m is the current frame number and j is the frequency point number; This represents the complex spectrum of the filtered air conduction signal at the j-th frequency point in the m-th frame after frequency domain transformation.

[0054] Step 3, Bone conduction signal strength and frequency domain calculation: The preprocessed framed bone conduction time domain signal is sent to the bone conduction signal strength detection unit to calculate the bone conduction signal probability. At the same time, the frequency domain transformation is performed on the signal, and the frequency domain energy distribution sequence of the bone conduction signal is obtained through the frequency domain detection unit. Compared with analyzing only the single time domain information, the frequency domain transformation of the signal can obtain more frequency domain feature information, which is convenient for analyzing the energy part of the signal in different frequency bands. In step 3, the specific process of calculating the bone conduction signal probability is as follows: Based on the framed bone conduction time domain signal after secondary filtering, the current frame power, the logarithmic weighted average power of the historical frames and the smoothed power value are calculated in sequence, and the bone conduction signal probability is updated frame by frame. Among them, the current frame power of the framed bone conduction time domain signal is calculated as shown in formula (7): (7);

[0055] in, This represents the bone conduction time-domain signal power in the m-th frame; The value represents the square of the nth sample point of the filtered bone conduction signal; the log-weighted average power of historical frames is calculated as shown in formula (8): (8);

[0056] in, This represents the log-weighted average power of the bone conduction signal across historical frames. Let represent the bone conduction time-domain signal power of the nth frame; input the squared value of the logarithmic weighted average power of the historical frames of the framed bone conduction time-domain signal and the logarithmic weighted average power of the historical frames into a first-order infinite impulse response filter to obtain the smoothed power value, as shown in formulas (9) and (10): (9);

[0057] (10);

[0058] in, This represents the log-weighted average power of the smoothed bone conduction signal across historical frames. This represents the log-weighted average power squared value of the bone conduction signal from historical frames after smoothing. This represents the smoothing coefficient used to calculate the log-weighted average power of the bone conduction signal across historical frames. Values: 63 / 64; This represents the smoothing coefficient used to calculate the log-weighted average squared power value of the bone conduction signal across historical frames. The value is 3 / 4; the bone conduction signal probability is obtained by updating the smooth power value frame by frame, as shown in formula (11): (11);

[0059] in, This represents the smoothing coefficient used to calculate the probability of bone conduction signals. The value is 7 / 8; in step 3, the frequency domain information of the bone conduction signal is obtained by performing a frequency domain transformation on the preprocessed framed bone conduction time domain signal, and the frequency domain power spectrum sequence of the bone conduction signal is calculated by combining the frequency domain transformation result, as shown in formula (12): (12);

[0060] in, This represents the frequency domain power spectrum sequence of the bone conduction signal; This represents the complex spectrum of the filtered bone conduction signal at the j-th frequency point in the m-th frame after frequency domain transformation.

[0061] Step 4, Gain Generation and Signal Output: The air conduction signal probability and bone conduction signal probability are combined to generate a frequency domain gain sequence. The frequency domain gain sequence is then clipped, limited, and smoothed to obtain the application gain. The application gain is applied to the air conduction frequency domain information and then subjected to inverse frequency domain transformation to output the enhanced time-domain air conduction signal. Compared with simple time-domain gain, frequency domain application gain can be applied to different frequency bands of the speech signal, and can more finely adjust the gain components of different frequencies to obtain a more accurate gain control effect. In step 4, the air conduction signal probability and bone conduction signal probability are combined to generate a frequency domain gain sequence. The frequency domain gain sequence is then clipped, limited, and smoothed to obtain the application gain. The application gain is applied to the air conduction frequency domain information and then subjected to inverse frequency domain transformation to output the enhanced time-domain air conduction signal. The specific steps are as follows: Step 4.1, Generate frequency domain gain as shown in formula (13): (13);

[0062] Where D represents the gain adjustment constant; Step 4.2, the frequency domain gain sequence is clipped and limited, as shown in formula (14): (14);

[0063] Step 4.3: Smooth the clipped and limited gain sequence as shown in formula (15): (15);

[0064] Step 4.4: Multiply the applied gain with the air conduction frequency domain information, perform an inverse frequency domain transform on the multiplication result, and finally output the enhanced time-domain air conduction signal after suppressing self-acoustic sound.

[0065] The present invention provides a self-sound detection and adjustment method for digital hearing aids based on air-bone conduction fusion. First, a first vibration sensor acquires a continuous air conduction signal. The air conduction signal is pre-processed using a first air conduction signal filter. The pre-processed framed air conduction time-domain signal is then sent to an air conduction signal intensity detection unit to calculate the air conduction signal probability. Simultaneously, a frequency domain transformation is performed on the signal to obtain the air conduction signal frequency domain information. Second, a second vibration sensor acquires a continuous bone conduction signal. This signal passes through a first bone conduction signal filter and a second bone conduction signal filter to obtain a filtered bone conduction time-domain signal. The pre-processed framed bone conduction time-domain signal is then sent to a bone conduction signal intensity detection unit to calculate the bone conduction signal probability. Simultaneously, a frequency domain transformation is performed on the signal, and a frequency domain detection unit obtains the bone conduction signal frequency domain energy distribution sequence. Finally, the... The probability of air conduction signal is combined with the probability of bone conduction signal to generate a frequency domain gain sequence. The frequency domain gain sequence is then clipped, limited, and smoothed to obtain the application gain. After the application gain is applied to the air conduction frequency domain information, it is transformed inversely to output an enhanced time-domain air conduction signal. Based on the air conduction microphone, this invention introduces a bone conduction vibration sensor to reduce the signal amplitude and compress the energy of the wearer's own voice when the wearer speaks. At the same time, when the input signal only contains the voices of others in the environment, the signal amplitude does not decrease and the voices of others are not suppressed. This allows the energy of the wearer's own voice to be adjusted to a reasonable level, improving the hearing experience when the wearer speaks. Compared with the traditional wide dynamic range compression algorithm in digital hearing aids, this invention can provide better self-sound detection and suppression effects.

[0066] To further illustrate the effects of the present invention, such as Figure 2 and Figure 3 The diagram shows the time-domain waveform of the input signal and the probability diagram of the wearer's own voice in the self-speech detection and adjustment method of the present invention. It is easy to observe that the top time-domain signal channel is the input signal collected by the air conduction microphone, which contains the wearer's own voice and the voices of others in the environment. The bottom time-domain signal channel is the probability of detecting the wearer's own voice. When only others in the environment are speaking, the probability of detecting self-speech remains at 0. When the wearer speaks, the probability of detecting self-speech can reflect the presence and intensity of their own voice in a timely manner, thereby achieving the purpose of detecting the wearer's own voice.

[0067] like Figures 4-7 The diagram shows the time-domain waveforms and spectrograms of the input signal and the output signal after self-voice adjustment in the self-voice detection and adjustment method of this invention. It is easy to observe that the top time-domain signal channel is the input signal collected by the air conduction microphone, containing the wearer's own speech and the voices of others in the environment. The bottom time-domain signal channel is the algorithm output signal after self-voice adjustment. The bottom shows a comparison of the spectrograms of the two channels. Figure 4and Figure 5 As can be seen, when the input signal only contains the voices of others in the environment, the signal amplitude does not decrease and the voices of others are not suppressed; when the wearer speaks, the signal amplitude of their own voice decreases and the energy is compressed, thereby adjusting the energy of the wearer's own voice to a reasonable level.

[0068] Furthermore, select an audio segment containing the speaker's own speech, and analyze its time-domain waveform and spectrogram, such as... Figures 8-11 As shown, the time-domain average RMS decreased by 11.17 dB, self-sound was significantly suppressed, and the gain at different frequencies decreased by 6-15 dB.

[0069] Furthermore, select an audio segment containing only spoken audio signals and obtain its time-domain waveform and spectrogram, such as... Figures 12-15 As shown, the average RMS attenuation in the time domain is less than 0.5dB, the voices of others are suppressed, and the gain attenuation at different frequencies is less than 0.5dB.

[0070] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A self-sound detection and adjustment method for digital hearing aids based on air-bone conduction fusion, characterized in that, Includes the following steps: Step 1, Signal Acquisition and Preprocessing: The air conduction continuous signal is collected by the first vibration sensor and the bone conduction continuous signal is collected by the second vibration sensor. The two continuous signals are framed and filtered preprocessed to obtain the preprocessed framed air conduction time domain signal and framed bone conduction time domain signal. Step 2, Calculation of air conduction signal strength and frequency domain: The preprocessed framed air conduction time domain signal is sent to the air conduction signal intensity detection unit to calculate the air conduction signal probability. At the same time, a frequency domain transformation is performed on the signal to obtain the air conduction signal frequency domain information. Step 3: Calculation of bone conduction signal intensity and frequency domain: The preprocessed framed bone conduction time-domain signal is sent to the bone conduction signal intensity detection unit to calculate the bone conduction signal probability. At the same time, frequency domain transformation is performed on the signal, and the frequency domain detection unit is used to obtain the frequency domain energy distribution sequence of the bone conduction signal. Step 4: Gain Generation and Signal Output The probabilities of air conduction signals and bone conduction signals are combined to generate a frequency domain gain sequence. The frequency domain gain sequence is then subjected to clipping, limiting, and smoothing processes to obtain the application gain. After applying the application gain to the air conduction frequency domain information, an inverse frequency domain transform is performed to output the enhanced time-domain air conduction signal. The specific steps are as follows: Step 4.1: Generate the frequency domain gain as shown in formula (13): （13）； Where D represents the gain adjustment constant; Indicates the probability of air conduction signal; Indicates the probability of bone conduction signal; This represents the frequency domain power spectrum sequence of the bone conduction signal; This represents the frequency domain power spectrum sequence of the air conduction signal; m represents the current frame number, and j represents the frequency point number; Step 4.2: Clip and limit the frequency domain gain sequence, as shown in formula (14): （14）； Step 4.3: Smooth the clipped and limited gain sequence as shown in formula (15): （15）； Step 4.4: Multiply the application gain with the air conduction frequency domain information, perform an inverse frequency domain transform on the multiplication result, and finally output the enhanced time-domain air conduction signal after suppressing self-acoustic sound.

2. The self-sound detection and adjustment method for a digital hearing aid based on air-bone conduction fusion according to claim 1, characterized in that: In step 1, the continuous air conduction signal collected by the first vibration sensor is divided into frames to obtain a framed air conduction signal. The framed air conduction signal is input into the first air conduction signal filter to obtain a framed air conduction time domain signal after removing the DC offset component. The continuous bone conduction signal acquired by the second vibration sensor is divided into frames to obtain a framed bone conduction signal. This framed bone conduction signal is first input into the first bone conduction signal filter to obtain a framed bone conduction signal with the DC offset component removed for the first time, and then input into the second bone conduction signal filter to obtain a framed bone conduction time domain signal with the high frequency component removed.

3. The self-sound detection and adjustment method for a digital hearing aid based on air-bone conduction fusion according to claim 1, characterized in that: In step 2, the specific process for calculating the probability of the air conduction signal is as follows: Step 2.1: Calculate the current frame power of the time-domain signal of the segmented air conduction, as shown in formula (1): （1）； in, This represents the air-conduction time-domain signal power in the m-th frame; n is the sample point number; N is a fixed frame length; This represents the square of the nth sample value of the filtered air conduction signal; Step 2.2: Calculate the log-weighted average power of historical frames, as shown in formula (2): （2）； in, This represents the log-weighted average power of the historical frames of the air conduction signal; K represents the number of historical frames selected; Step 2.3: Input the squared value of the logarithmic weighted average power of the historical frames of the segmented air conduction time domain signal and the logarithmic weighted average power of the historical frames into a first-order infinite impulse response filter to obtain the smoothed power value, as shown in formulas (3) and (4): （3）； （4）； in, This represents the log-weighted average squared power value of the historical frames of the smoothed air conduction signal; This represents the log-weighted average power of the smoothed air conduction signal across historical frames. This represents the smoothing coefficient used to calculate the logarithmically weighted average power of the historical frames of the air conduction signal. Values: 63 / 64; This represents the smoothing coefficient used to calculate the log-weighted average squared power value of the historical frames of the air conduction signal. The value is 3 / 4; Step 2.4: Update the air conduction signal probability frame by frame based on the smoothed power value, as shown in formula (5): （5）； This represents the smoothing coefficient used to calculate the probability of the air conduction signal. The value is 7 / 8.

4. The self-sound detection and adjustment method for a digital hearing aid based on air-bone conduction fusion according to claim 3, characterized in that: In step 2, the frequency domain information of the air conduction signal is obtained by performing a frequency domain transformation on the preprocessed framed air conduction time domain signal, and the frequency domain power spectrum sequence of the air conduction signal is calculated by combining the frequency domain transformation result, as shown in formula (6): （6）； This represents the complex spectrum of the filtered air conduction signal at the j-th frequency point in the m-th frame after frequency domain transformation.

5. The self-sound detection and adjustment method for a digital hearing aid based on air-bone conduction fusion according to claim 4, characterized in that: In step 3, the specific process of calculating the bone conduction signal probability is as follows: based on the framed bone conduction time domain signal after secondary filtering, the power of the current frame, the logarithmic weighted average power of the historical frames and the smoothed power value are calculated in sequence, and the bone conduction signal probability is updated frame by frame. The current frame power of the frame-guided time-domain signal is calculated as shown in formula (7): （7）； in, This represents the bone conduction time-domain signal power in the m-th frame; This represents the square of the nth sample value of the filtered bone conduction signal; The log-weighted average power of historical frames is calculated as shown in formula (8): （8）； in, This represents the log-weighted average power of the bone conduction signal across historical frames. This represents the bone conduction time-domain signal power of the nth frame; The squared value of the logarithmic weighted average power of the historical frames of the framed bone conduction time-domain signal and the logarithmic weighted average power of the historical frames are respectively input into a first-order infinite impulse response filter to obtain the smoothed power value, as shown in formulas (9) and (10): （9）； （10）； in, This represents the log-weighted average power of the smoothed bone conduction signal across historical frames. This represents the log-weighted average power squared value of the bone conduction signal from historical frames after smoothing. This represents the smoothing coefficient used to calculate the log-weighted average power of the bone conduction signal across historical frames. Values: 63 / 64; This represents the smoothing coefficient used to calculate the log-weighted average squared power value of the bone conduction signal across historical frames. The value is 3 / 4; The probability of bone conduction signal is obtained by updating the smooth power value frame by frame, as shown in formula (11): （11）； in, This represents the smoothing coefficient used to calculate the probability of bone conduction signals. The value is 7 / 8.

6. The self-sound detection and adjustment method for a digital hearing aid based on air-bone conduction fusion according to claim 5, characterized in that: In step 3, the frequency domain information of the bone conduction signal is obtained by performing a frequency domain transformation on the preprocessed framed bone conduction time domain signal, and the frequency domain power spectrum sequence of the bone conduction signal is calculated by combining the frequency domain transformation result, as shown in formula (12): （12）； in, This represents the complex spectrum of the filtered bone conduction signal at the j-th frequency point in the m-th frame after frequency domain transformation.

Images