Signal processing method of auditory assistance device and related product

Abstract

The invention provides a signal processing method of an auditory assistance device and a related product. According to the signal processing method of the auditory auxiliary device, the auditory auxiliary device comprises a microphone, a receiver, a signal processing module arranged between the microphone and the receiver, and an adaptive filtering module connected with the signal processing module, and the signal processing method comprises the following steps: obtaining a superposed signal input by the microphone, the superposed signal is formed by superposing an original signal and a feedback signal obtained after an output signal of the receiver passes through the feedback path; filtering at least a part of linear feedback part in the signal processed by the signal processing module by using an adaptive filtering module to obtain a real-time error signal; and taking the real-time error signal as the input of a preset neural network and the original signal as the target of the neural network, and training the preset neural network to obtain a neural network model for estimating the residual signal of the adaptive filtering. According to the scheme, the feedback suppression performance and the overall sound quality can be effectively improved.

Description

Technical Field

[0001] This invention relates to signal processing of acoustic devices, and in particular to a signal processing method and related products for a hearing aid. Background Technology

[0002] Hearing assistive devices are electronic devices designed to improve the auditory experience of hearing-impaired individuals and assist them in perceiving sound information. Typical products include hearing aids, hearing-aided headphones, and sound reinforcement systems. Their core working principle is as follows: a microphone collects sound signals from the environment (i.e., the raw signal), which are then processed by a signal processing module through gain adjustment, noise reduction, and frequency compensation. Finally, a receiver converts the processed signal into an acoustic signal that can be perceived by hearing-impaired individuals, thereby compensating for hearing loss. The miniaturization and portability requirements of these devices (the microphone and receiver are usually close together) lead to severe feedback howling problems in practical use, becoming a core technical bottleneck restricting product performance and user experience.

[0003] During the operation of a hearing aid, the acoustic signal output from the receiver is propagated back to the microphone input through a feedback path formed by the acoustic path inside the ear canal and reflections from the surrounding environment. At this point, the microphone does not simply capture the original signal, but rather a superimposed signal formed by the original signal and the feedback signal. Under high-gain conditions, this easily satisfies the positive feedback condition. That is, when the feedback signal and feedback path meet specific conditions, it can trigger self-excited oscillation of the system, producing sharp howling noise. This howling not only severely interferes with the clarity of the target sound, but may also damage the user's residual hearing due to high-frequency, strong noise stimulation, and may even cause the device to malfunction.

[0004] To address the feedback howling problem, the most common solution in existing technologies is to use an adaptive filtering module to suppress the feedback signal. The core idea is to estimate the transfer function of the feedback path using an adaptive algorithm, generating a cancellation signal with a similar amplitude but opposite phase to the feedback signal, thereby filtering out the linear feedback component. However, in practical applications: in scenarios where the feedback path changes rapidly, the adaptive filter's tracking speed struggles to keep up, and increasing the tracking speed leads to greater speech distortion; in scenarios with strongly correlated signals (such as music), the adaptive filter's convergence deviates from the true path, causing speech distortion. Existing hearing aids struggle to meet the requirements of fast convergence speed, high gain, and consistent sound quality.

[0005] As hearing-impaired individuals increasingly demand higher sound quality from assistive devices, and as the application scenarios for these devices become more diverse (such as noisy public places and sports scenarios), the limitations of traditional adaptive filtering schemes are becoming more and more prominent. There is an urgent need for a signal processing technology that can accurately suppress linear feedback while effectively handling residual feedback in special scenarios. Summary of the Invention

[0006] One object of the present invention is to provide a method to improve the feedback suppression performance of hearing aids.

[0007] A further objective of this invention is to enhance the robustness of the adaptive filtering algorithm.

[0008] A further objective of this invention is to reduce the computational resources required for signal processing in hearing aids.

[0009] Specifically, the present invention provides a signal processing method for a hearing aid device, the hearing aid device including a microphone, a receiver, a signal processing module disposed between the microphone and the receiver, and an adaptive filtering module connected to the signal processing module, and the signal processing method including: The superimposed signal input from the microphone is acquired. The superimposed signal is formed by superimposing the original signal and the feedback signal after the receiver's output signal passes through the feedback path. An adaptive filtering module is used to filter out at least a portion of the linear feedback component in the signal processed by the signal processing module in order to obtain a real-time error signal. The real-time error signal is used as the input to a pre-defined neural network, and the original signal is used as the target of the neural network. The pre-defined neural network is trained to obtain a neural network model for estimating the residual signal of adaptive filtering.

[0010] Optionally, the step of acquiring the superimposed signal collected by the microphone includes: Identify multiple feedback paths; For each feedback path, the signal output by the receiver after the hearing aid device processes the original signal is simulated to obtain the feedback signal corresponding to the original signal. The original signal is superimposed with the feedback signal to obtain the superimposed signal.

[0011] Optionally, the steps for determining multiple feedback paths include: Collect measured feedback paths from multiple typical application scenarios; Based on the measured feedback paths, a preset number of simulated feedback paths are generated through simulation. The measured feedback path and the simulation feedback path are used as the feedback path.

[0012] Optionally, after obtaining the neural network model, the following may also be included: An inference module is placed inside the hearing aid device; the inference module is used to deploy a neural network model. During the use of the hearing aid, the inference module uses a neural network model to infer the signal after it has been filtered by the adaptive filtering module, thereby obtaining the desired signal for the signal processing module to continue processing.

[0013] Optionally, the neural network model includes: The feature extraction unit is configured to extract features of the signal after it has been filtered by the adaptive filtering module. The encoding unit is configured to encode the features extracted by the feature extraction unit; The network processing unit is configured to perform network processing on the feature codes to obtain the features of the desired signal; The decoding unit is configured to decode the features of the desired signal to obtain the desired signal.

[0014] Optionally, the encoding unit adopts a stacked structure of convolutional layers, batch normalization layers, and activation function layers; The decoding unit adopts a stacked structure of deconvolutional layers, batch normalization layers, and activation function layers.

[0015] Optionally, the feature extraction unit is configured to extract the short-time Fourier transform frequency domain features and / or Mel frequency cepstral coefficient features of the signal.

[0016] According to another aspect of the present invention, a hearing aid is also provided. The hearing aid includes: A microphone, configured to receive sound signals; The receiver is configured to output audio signals; The signal processing module is located between the microphone and the receiver and is configured to process audio signals. An adaptive filtering module is connected to the signal processing module and configured to filter out at least a portion of the linear feedback portion of the signal processed by the signal processing module. The inference module is used to deploy the neural network model and is configured to infer the signal after it has been filtered by the adaptive filtering module, so as to obtain the desired signal for further processing by the signal processing module; the neural network model is obtained through the signal processing method of any of the above-mentioned hearing aids.

[0017] Optionally, the neural network model includes: The feature extraction unit is configured to extract features of the signal after it has been filtered by the adaptive filtering module. The encoding unit is configured to encode the features extracted by the feature extraction unit; The network processing unit is configured to perform network processing on the feature codes to obtain the features of the desired signal; The decoding unit is configured to decode the features of the desired signal to obtain the desired signal; and The encoding unit adopts a stacked structure of convolutional layers, batch normalization layers, and activation function layers; The decoding unit adopts a stacked structure of deconvolutional layers, batch normalization layers, and activation function layers; The feature extraction unit is configured to extract the frequency domain features and / or Mel frequency cepstral coefficient features of the signal by performing a short-time Fourier transform.

[0018] According to another aspect of the present invention, a computer device is also provided. The computer device includes a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the signal processing method for any of the aforementioned hearing assistive devices.

[0019] The signal processing method of the hearing aid device of the present invention acquires a superimposed signal containing the original signal and the feedback signal. First, an adaptive filtering module is used to selectively filter out some linear feedback components, achieving initial suppression of linear feedback and reducing the basic interference of feedback howling. Then, a neural network model is trained using the filtered real-time error signal as input and the original signal as the target. This enables the model to accurately learn the patterns of the filter residual signal, providing support for the hearing aid device to use the target-trained neural network model for anomaly handling. This scheme combines the linear feedback component processing of adaptive filtering with the powerful fitting ability of neural networks, leveraging the efficiency of adaptive filtering in handling linear feedback while ensuring the accuracy of the model's estimation of residual signals through targeted training, laying the foundation for improving the overall feedback suppression effect.

[0020] Furthermore, the hearing aid device and its signal processing method of the present invention integrate adaptive filtering and deep learning to realize a hybrid feedback suppression method. This scheme utilizes an adaptive filter to dynamically estimate the feedback path, fully leveraging its advantages in real-time performance and low complexity in processing linear feedback components; and employs a deep neural network to quickly predict and process the feedback signal remaining after filtering in special scenarios. The collaborative work of adaptive filtering and deep learning effectively improves feedback suppression performance and overall sound quality while maintaining low system latency and low power consumption.

[0021] Furthermore, the hearing aid device and signal processing method of the present invention can completely suppress feedback howling, improve the clarity and fidelity of the target sound, and improve the auditory experience of hearing-impaired people; and expand the applicability of the device in complex acoustic environments.

[0022] The above and other objects, advantages and features of the present invention will become more apparent to those skilled in the art from the following detailed description of specific embodiments of the invention in conjunction with the accompanying drawings. Attached Figure Description

[0023] The following sections will describe some specific embodiments of the invention in detail by way of example and not limitation, with reference to the accompanying drawings. The same reference numerals in the drawings denote the same or similar parts or portions. Those skilled in the art should understand that these drawings are not necessarily drawn to scale. In the drawings: Figure 1 This is a schematic diagram of the neural network training process of a hearing aid device according to an embodiment of the present invention; Figure 2 This is a schematic diagram of a hearing aid device according to an embodiment of the present invention using a neural network model for reasoning; Figure 3 This is a schematic block diagram of a neural network model in a hearing aid device according to an embodiment of the present invention. Figure 4 This is a schematic diagram of a signal processing method for a hearing aid device according to an embodiment of the present invention; Figure 5 This is a room impulse response diagram of the original signal and the desired signal in the signal processing method of a hearing aid device according to an embodiment of the present invention; Figure 6 This is a schematic diagram of the time and frequency domains of the measured feedback path in the signal processing method of a hearing aid device according to an embodiment of the present invention; Figure 7 This is a schematic diagram illustrating how whistling is generated in a signal processing method for a hearing aid device according to an embodiment of the present invention; Figure 8 This is a diagram showing the effect of signal processing by a signal processing method for a hearing aid device according to an embodiment of the present invention; Figure 9 This is a schematic block diagram of a computer device according to an embodiment of the present invention. Detailed Implementation

[0024] Figure 1 This is a schematic diagram of a neural network training process performed by a hearing aid device according to an embodiment of the present invention. The hearing aid device may include: a microphone 101, a receiver 102, a signal processing module 103 disposed between the microphone 101 and the receiver 102, and an adaptive filtering module 104 connected to the signal processing module 103.

[0025] The microphone 101 (also known as a microphone or mic) is the sound acquisition component of a hearing aid device, used to convert external mechanical sound wave signals (sound signals) into electrical signals that the device can process.

[0026] The receiver 102 (also known as the earpiece or speaker) is the sound output component of the hearing aid device, used to convert the processed electrical signal back into a mechanical sound wave signal (sound) and transmit it into the user's ear canal.

[0027] The signal processing module 103 is a signal processing unit connecting the microphone 101 and the receiver 102. It processes the raw electrical signal input from the microphone to adapt it to the user's hearing needs. The signal processing module 103 can perform functions such as wide dynamic range compression (WDRC), directional optimization, and signal amplification. The signal processing module 103 can use a digital signal processor (DSP), along with a storage chip (for algorithms and various parameters) and a control chip (for receiving user input), to achieve the corresponding functions.

[0028] The adaptive filtering module 104 is connected to the signal processing module 103 and dynamically adjusts the filtering parameters according to changes in ambient noise. The adaptive filtering module 104 can use either the Normalized Least Mean Squares (NLMS) algorithm or the Least Mean Squares (LMS) algorithm.

[0029] The signal flow in the hearing aid device is as follows: the original signal X(n) and the feedback signal after passing through feedback path 111 are superimposed at microphone 101. The superimposed signal is acquired by microphone 101 and converted into a corresponding electrical signal. The adaptive filtering module adaptively filters the signal converted by microphone 101 to remove at least a portion of the linear feedback part v(n) to obtain the real-time error signal, i.e., signal e(n). Signal e(n) contains the desired signal and the residual signal of the adaptive feedback. Signal processing module 103 processes signal e(n) to obtain signal u(n), which is then converted back into sound signal y(n) by receiver 102. The sound signal y(n) is superimposed with the original signal X(n) at microphone 101 after passing through feedback path 111.

[0030] In the neural network training process of the hearing aid device in this embodiment, a large number of feedback paths can be obtained through actual measurement and simulation, and feedback signals can be generated for these paths. The real-time error signal e(n) is used as the input of the preset neural network, while the original signal X(n) is used as the target of the neural network to train the preset neural network. Through deep learning, the trained neural network model can estimate the residual signal of the adaptive feedback by utilizing the network's powerful fitting ability, thereby outputting the final desired signal.

[0031] Figure 2This is a schematic diagram illustrating the use of a neural network model for inference in a hearing aid device according to an embodiment of the present invention. The inference stage refers to the stage in which the hearing aid device actually uses the neural network model. The trained neural network model is deployed in the inference module 122. The input signal of the neural network model is the signal processed in real time by the adaptive filtering module 104, i.e., signal e(n), and the output is the desired signal after further removing residual feedback. The desired signal is further processed by the signal processing module 103 to perform WDRC, directivity optimization, etc. By further removing the residual feedback signal in the adaptive feedback part, the effect of adaptive filtering is improved and the sound quality is enhanced, the robustness of the adaptive filtering algorithm is strengthened, so that the hearing aid device can operate in a wider gain range without howling, thus improving the overall sound quality.

[0032] The hearing aid device in this embodiment utilizes the adaptive filtering module 104 to dynamically estimate the feedback path, fully leveraging its advantages in real-time performance and low complexity in handling linear feedback components. Furthermore, the inference module 122 incorporates a deep neural network model to rapidly predict and suppress complex feedback components in specific scenarios. The collaborative work of the adaptive filtering module 104 and the deep neural network model of the inference module 122 effectively improves feedback suppression performance and overall sound quality while maintaining low system latency and low power consumption. This hearing aid device provides users with stronger sound amplification without generating howling, making it particularly suitable for users with moderate to severe hearing loss. Effective feedback control avoids harsh high-frequency howling interference, resulting in a more natural and gentle sound output from the hearing aid device, improving speech intelligibility and overall auditory comfort. The hearing aid device is suitable for various complex wearing environments, such as telephone calls, close proximity to reflective surfaces, and wearing hats. Through the collaborative work of adaptive filtering and deep neural networks, it can more accurately estimate the feedback path and effectively suppress linear and residual feedback components, significantly reducing the probability and intensity of howling. It effectively reduces sudden howling caused by environmental changes, improving product stability and user satisfaction.

[0033] In the embodiment employing the NLMS algorithm, the adaptive filtering module 104 estimates the feedback path under the criterion of minimizing the mean square error, thus obtaining the signal after removing linear feedback. The output signal s(n) of the microphone 101 can be expressed as: Updated via NLMS algorithm Thus obtain Here, w(n) is the coefficient vector of the adaptive filter, used to estimate the feedback path. wT(n)s(n) represents the estimated value of the feedback sound by the adaptive filter, which can be denoted as v(n). That is, the input signal s(n) is linearly weighted by the coefficients w(n) to simulate the linear component of the feedback signal. μ is the step size parameter, representing the interval from n to n+1. By continuously updating w(n), the signal more closely approximates the linear characteristics of the real feedback path. The final error signal e(n) approximates the signal after removing the linear feedback, which contains the residual feedback component and the desired signal. This part of the signal will be further processed by the neural network model to further improve the speech quality and the robustness of the algorithm.

[0034] Figure 3 This is a schematic block diagram of a neural network model in a hearing aid device according to an embodiment of the present invention. The neural network model may include: a feature extraction unit 131, an encoding unit 132, a network processing unit 133, and a decoding unit 134.

[0035] Feature extraction unit 131 is configured to extract features from the signal filtered by adaptive filtering module 104. Encoding unit 132 is configured to encode the features extracted by feature extraction unit 131. Network processing unit 133 is configured to perform network processing on the encoded features to obtain features of the desired signal. Decoding unit 132 is configured to decode the features of the desired signal to obtain the desired signal.

[0036] The feature extraction unit 131 can be configured to extract the frequency domain features of the short-time Fourier transform (STFT) of the signal and / or the cepstral coefficient features of the Mel frequency (Mel). STFT divides the time-domain signal into short segments, performs a Fourier transform on each segment, and the resulting frequency domain features can clearly identify the frequency peaks of the howling, facilitating the model's localization of the howling components that need to be suppressed. Mel features, based on human auditory characteristics, map the linear frequency axis to the Mel scale, resulting in features that better match human auditory perception, more accurately reducing redundant information and improving model efficiency.

[0037] The aforementioned encoding unit 132 can employ a stacked structure of convolutional layers (conv), batch normalization layers (batch normalization), and activation function layers (activation), i.e., a Bottleneck structure. The convolutional layers extract local time-frequency features through convolutional kernels; the batch normalization layers normalize the convolutional outputs, accelerating model training and avoiding training instability caused by fluctuations in input distribution. The activation function layers introduce non-linearity, enabling the model to learn complex feedback suppression patterns. After multiple stacks of convolutions, the feature dimension is compressed to a lower dimension, allowing the model to retain only the most critical features distinguishing between the howling and desired signals. This reduces computational load, thus adapting to the low-power requirements of hearing aids, and also filters redundant information, improving the model's ability to learn core patterns. The decoding unit 134 can employ a stacked structure of deconvolutional layers (deconv), batch normalization layers (batch normalization), and activation function layers (activation). The deconvolutional layers are symmetrical to the convolutional layers, gradually expanding the feature dimension through deconvolution operations to restore it to a dimension matching the desired signal.

[0038] To ensure convergence and final results, neural network models can also introduce skip operations. By skipping, the original input features are introduced, reducing the neural network model's dependence on local noise features generated by intermediate processing layers and avoiding overfitting to local noise in the training data.

[0039] Figure 4 This is a schematic diagram of a signal processing method for a hearing aid device according to an embodiment of the present invention. The signal processing method for the hearing aid device generally includes: Step S401: Obtain the superimposed signal input from the microphone. The superimposed signal is formed by superimposing the original signal and the feedback signal after the output signal of the receiver passes through the feedback path.

[0040] This step may include: determining multiple feedback paths; for each feedback path, performing feedback simulation on the signal output by the receiver after the hearing aid device processes the original signal to obtain a feedback signal corresponding to the original signal; and superimposing the original signal and the feedback signal to obtain a superimposed signal.

[0041] Feedback paths can include measured feedback paths and simulated feedback paths. In some embodiments, the step of determining multiple feedback paths may include: collecting measured feedback paths from multiple typical application scenarios; generating a preset number of simulated feedback paths based on the measured feedback paths; and using the measured feedback paths and simulated feedback paths as feedback paths. A single feedback path has limited coverage and cannot adapt to all complex scenarios in the actual use of hearing aids (such as different wearing states and changes in ambient acoustics). The solution in this embodiment expands the coverage dimension of feedback paths by combining measured feedback paths and simulated feedback paths, thus solving the problem of scenario limitations. By adapting to multiple feedback paths and leveraging the generalization ability of neural networks, the stability of feedback suppression under different scenarios is improved. Generating a preset number of simulated feedback paths based on measured feedback paths makes the simulated feedback paths closer to actual use scenarios, simplifying the training process to some extent.

[0042] In real-world scenarios where sudden path changes (such as making or receiving a phone call) cause a sudden howling, this embodiment incorporates simulations of such path changes, i.e., a sudden alteration of the feedback path. If the positive feedback condition is met. The system may be in an unstable state, causing the howling sound to continuously increase. For the transmission gain of hearing aids, (For feedback transfer function). Simulation part while ensuring sound quality. The attenuation of target data allows the model to learn the conditions that disrupt feedback, thus ensuring in principle that the sound will not produce a howling effect.

[0043] Step S402: Use the adaptive filtering module to filter out at least a portion of the linear feedback portion of the signal processed by the signal processing module to obtain the real-time error signal.

[0044] Step S403: The real-time error signal is used as the input of the preset neural network and the original signal is used as the target of the neural network to train the preset neural network, so as to obtain a neural network model for estimating the residual signal of adaptive filtering.

[0045] The signal processing method of the hearing aid device of the present invention acquires a superimposed signal containing the original signal and the feedback signal. First, an adaptive filtering module is used to selectively filter out some linear feedback components, achieving initial suppression of linear feedback and reducing the basic interference of feedback howling. Then, a neural network model is trained using the filtered real-time error signal as input and the original signal as the target. This enables the model to accurately learn the patterns of the filter residual signal, providing support for the hearing aid device to use the target-trained neural network model for abnormal situation handling. This scheme combines the linear feedback component processing of adaptive filtering with the powerful fitting ability of neural networks, leveraging the efficiency of adaptive filtering in handling linear feedback while ensuring the accuracy of the model's estimation of residual signals through targeted training, laying the foundation for improving the overall feedback suppression effect.

[0046] After obtaining the neural network model, the method in this embodiment deploys an inference module within the hearing aid device. This inference module is used to deploy the neural network model. During use of the hearing aid device, the inference module uses the neural network model to infer the signal filtered by the adaptive filtering module, thereby obtaining the desired signal for further processing by the signal processing module. The neural network model realizes the implementation of the model from training to real-time inference.

[0047] Figure 5 This is a room impulse response diagram of the original signal and the desired signal in the signal processing method of a hearing aid device according to an embodiment of the present invention. In the diagram, the horizontal axis represents time in seconds (Fs); the vertical axis represents severity. The original room impulse response during speech is Rir1; the desired room impulse response is Rir2. Simulations were conducted to... Make This generates data that partially disrupts the positive feedback conditions, enabling the network model to disrupt the positive feedback conditions and thus block the formation of feedback.

[0048] Figure 6 This is a schematic diagram of the time and frequency domains of the measured feedback path in the signal processing method of a hearing aid device according to an embodiment of the present invention. Figure 7 This is a schematic diagram illustrating how whistling is generated in a signal processing method for a hearing aid device according to an embodiment of the present invention; Figure 8 This is a diagram illustrating the effect of signal processing by a hearing aid device according to an embodiment of the present invention. When the gain of the hearing aid device is 0dB, the system gain is just in a stable state. When the system gain increases beyond 0dB, the system will be in an unstable state, initially producing howling (e.g., whistling) around 3200Hz and 4500Hz. Figure 7(As shown). At this time, the system is in a positive feedback state, and the system will continuously howl, seriously affecting the user experience. The method in this embodiment first uses adaptive filtering to suppress the feedback signal, and then uses a neural network model to process the residual feedback signal, by comparison. Figure 8 as well as Figure 9 All the frequencies that originally produced feedback no longer produced feedback, and the voice quality was improved, enhancing the user experience.

[0049] This embodiment also provides a computer device 830. Figure 9 This is a schematic block diagram of a computer device 830 according to an embodiment of the present invention. The computer device 830 may include a memory 832, a processor 831, and a computer program 811 stored in the memory 832 and running on the processor 831. When the computer program 811 is executed by the processor 831, it implements the steps of the signal processing method of the hearing assistive device of any of the above embodiments, and completes the simulation of the feedback path, model training, and the actual application process of the hearing assistive device.

[0050] The computer program 811 used to perform the operations of this invention may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-related instructions, microcode, firmware instructions, status setting data, integrated circuit configuration data, or source code or object code written in any combination of one or more programming languages ​​and procedural programming languages.

[0051] Computer program 811 may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the latter case, the remote computer may be connected to the user's computer via any type of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, to perform aspects of the invention, electronic circuits including, for example, programmable logic circuits, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs) may execute computer-readable program instructions to personalize the electronic circuits by utilizing state information of computer-readable program instructions.

[0052] Therefore, those skilled in the art should recognize that although numerous exemplary embodiments of the present invention have been shown and described in detail herein, many other variations or modifications conforming to the principles of the present invention can be directly determined or derived from the disclosure of the present invention without departing from the spirit and scope of the invention. Thus, the scope of the present invention should be understood and construed as covering all such other variations or modifications.

Claims

1. A signal processing method of a hearing assistance device, wherein the hearing assistance device comprises a microphone, a receiver, a signal processing module arranged between the microphone and the receiver, and an adaptive filter module connected to the signal processing module, characterized in that The signal processing method comprises: acquiring a superimposed signal input by the microphone, the superimposed signal being formed by superimposing an original signal and a feedback signal after the output signal of the microphone via a feedback path; using the adaptive filtering module to filter at least part of a linear feedback part in the signal processed by the signal processing module to obtain a real-time error signal; training a preset neural network by taking the real-time error signal as an input of the preset neural network and the original signal as a target of the neural network, to obtain a neural network model for estimating a residual signal of the adaptive filtering.

2. The signal processing method of an auditory assistive device according to claim 1, wherein, The step of acquiring the superimposed signal collected by the microphone comprises: determining a plurality of feedback paths; for each of the feedback paths, performing feedback simulation on a signal output by the microphone after the original signal is processed by the hearing aid device to obtain a feedback signal corresponding to the original signal; superimposing the original signal and the feedback signal to obtain the superimposed signal.

3. The signal processing method of an auditory assistive device according to claim 2, wherein, The step of determining a plurality of feedback paths comprises: collecting a plurality of measured feedback paths of typical application scenarios; generating a preset number of simulated feedback paths by simulation on the basis of the measured feedback paths; taking the measured feedback paths and the simulated feedback paths as the feedback paths.

4. The signal processing method of an auditory assistive device according to claim 1, wherein, After obtaining the neural network model, further comprising: arranging an inference module in the hearing aid device, the inference module being configured to deploy the neural network model; in the use process of the hearing aid device, using the neural network model by the inference module to infer a signal filtered by the adaptive filtering module, thereby obtaining a desired signal for the signal processing module to continue processing.

5. The signal processing method of an auditory assistive device according to claim 4, wherein, The neural network model comprises: a feature extraction unit configured to extract features of a signal filtered by the adaptive filtering module; an encoding unit configured to encode the features extracted by the feature extraction unit; a network processing unit configured to perform network processing on the feature encoding to obtain features of a desired signal; a decoding unit configured to decode the features of the desired signal to obtain the desired signal.

6. The signal processing method of the hearing aid device according to claim 5, wherein the encoding unit adopts a stacked structure of a convolutional layer, a batch normalization layer and an activation function layer; the decoding unit adopts a stacked structure of a deconvolutional layer, a batch normalization layer and an activation function layer.

7. The signal processing method of the hearing aid device according to claim 5, wherein the feature extraction unit is configured to extract short-time Fourier transform frequency domain features and / or mel-frequency cepstral coefficient features of a signal.

8. A hearing assistance device, characterized in that comprises: a microphone configured to receive a sound signal; a microphone configured to output a sound signal; a signal processing module arranged between the microphone and the microphone and configured to process a sound signal; an adaptive filtering module connected with the signal processing module and configured to filter at least part of a linear feedback part in the signal processed by the signal processing module; an inference module configured to deploy the neural network model and configured to perform inference on the signal filtered by the adaptive filter module to obtain a desired signal for further processing by the signal processing module; the neural network model is obtained by performing the signal processing method of the hearing aid device of any one of claims 1 to 7.

9. The hearing assistance device of claim 8, wherein, the neural network model comprises: a feature extraction unit configured to extract features of the signal filtered by the adaptive filter module; an encoding unit configured to encode the features extracted by the feature extraction unit; a network processing unit configured to perform network processing on the feature encoding to obtain features of the desired signal; a decoding unit configured to decode the features of the desired signal to obtain the desired signal; and the encoding unit employs a stack structure of a convolution layer, a batch normalization layer, and an activation function layer; the decoding unit employs a stack structure of a deconvolution layer, a batch normalization layer, and an activation function layer; the feature extraction unit is configured to extract frequency domain features of short-time Fourier transform and / or mel-frequency cepstral coefficient features of the signal.

10. A computer device comprising a memory, a processor, and a computer program stored on the memory, wherein the computer program comprises instructions that, when executed by the processor, cause the processor to perform the method of any one of claims 1-9. the processor executes the computer program to implement the steps of the signal processing method of the hearing aid device of any one of claims 1 to 7.

Images