Conduction device fitting management system and conduction device

Abstract

The invention discloses a conduction device fitting management system and a conduction device, and belongs to the technical field of data analysis, and the method specifically comprises the steps: collecting multi-source fitting data, carrying out the real-time fusion, and constructing a scene-physiology-loss three-dimensional self-adaptive model through the combination of a regional fitting strategy defined based on a geographic position or an environment type; according to the real-time fusion data and model data, dynamically allocating three types of data priorities and generating initial compensation parameters, inputting the initial compensation parameters into a third-order LSTM acoustic compensation engine containing an attention mechanism, and outputting acoustic compensation signals in combination with hearing audio signals collected by a real-time hearing test strategy; converting the acoustic compensation signal into a hearing aid DSP executable instruction, driving the hearing aid to compensate and output a gain curve, and collecting user feedback at the same time; the system can adaptively adjust the compensation parameters according to the real-time state of the user and the environment change, and improves the compensation effect of the hearing aid.

Description

Technical Field

[0001] The present invention belongs to the technical field of data analysis, and in particular to a conduction device fitting management system and a conduction device. Background Art

[0002] With the continuous development of science and technology, transducers are playing an increasingly important role in people's lives, especially in hearing assistance, communications, and other fields. The main function of a transducer is to convert sound signals into electrical signals and transmit them to the user through vibration transmission, allowing the user to perceive the sound. However, in actual application, the fitting and acoustic compensation effect of the transducer directly affect the user experience.

[0003] Traditional methods of fitting conduction devices typically rely on professionals testing and adjusting the user in specific environments. This approach has many limitations. First, the test environment is often relatively simple and cannot simulate the various complex scenarios that users may encounter in real life, such as noisy public places or quiet indoor environments. Therefore, fitting based on the results of single-environment testing may not meet the user's needs in different scenarios in actual use, resulting in significant differences in the user's hearing effect in different environments.

[0004] Secondly, traditional fitting methods mainly focus on physiological indicators such as the degree of hearing loss, while ignoring changes in the user's physiological state and the wear and tear of the conduction device itself. The user's physiological state will change over time, physical condition, and other factors. For example, fatigue and mood swings may affect the user's perception of sound. At the same time, the conduction device will gradually experience problems such as component wear and aging during use. These losses will also affect the performance of the device and the acoustic compensation effect. Traditional fitting methods cannot adapt to these changes in a timely manner, resulting in limited accuracy and effectiveness of fitting results. Therefore, it is of great practical significance to develop a conduction device fitting management system that can comprehensively consider multi-source fitting data, realize dynamic compensation parameter generation and real-time feedback. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention proposes a conduction device fitting management system and a conduction device, which collects multi-source fitting data and fuses them in real time, and combines regional fitting strategies defined based on geographic location or environment type to construct a three-dimensional adaptive model of scene-physiology-loss; based on real-time fusion data and model data, dynamically allocates priorities of three types of data and generates initial compensation parameters, which are input into a three-order LSTM acoustic compensation engine with an attention mechanism, and outputs an acoustic compensation signal in combination with the hearing audio signal collected by the real-time hearing test strategy; the acoustic compensation signal is converted into an executable instruction of the hearing aid DSP, drives the hearing aid compensation and outputs a gain curve, and collects user feedback at the same time; the system can adaptively adjust the compensation parameters according to the user's real-time status and environmental changes, thereby improving the hearing aid compensation effect.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A conduction device fitting management system, comprising: a data acquisition module, a three-dimensional modeling module, a compensation parameter generation module, an acoustic compensation module, and an execution and feedback module;

[0008] The three-dimensional modeling module performs real-time fusion of the multi-source fitting data collected by the data acquisition module and constructs a scene-physiology-loss three-dimensional adaptive model in combination with a pre-defined regional fitting strategy;

[0009] The compensation parameter generation module dynamically assigns priorities to the three types of data based on the real-time fused multi-source fitting data and the scene-physiology-loss three-dimensional adaptive model data, and generates initial compensation parameters;

[0010] Inputting the initial compensation parameters into a third-order LSTM acoustic compensation engine configured in the acoustic compensation module, combining the audio signal collected according to a predefined real-time hearing test strategy, outputting an acoustic compensation signal, and transmitting it to the execution and feedback module;

[0011] The execution and feedback module converts the acoustic compensation signal into executable instructions for the main control chip of the conduction device, drives the conduction device to perform acoustic compensation, and outputs the gain curve in real time. At the same time, it collects user feedback information on the current acoustic compensation effect.

[0012] Specifically, the three-dimensional modeling module performs real-time fusion of the multi-source fitting data collected by the data acquisition module, and combines it with a pre-defined regional fitting strategy to construct a scene-physiology-loss three-dimensional adaptive model, including:

[0013] A1: Collect multi-source fitting data; the multi-source fitting data includes environmental acoustic feature data, physiological index data, and motion data;

[0014] A2: defining a regionalized fitting strategy; the regionalized fitting strategy includes defining an environmental scene category set, a physiological state category set, and a loss feature set;

[0015] A3: Establish a three-dimensional strategy matrix based on the environmental scenario category, physiological state category, and loss characteristics. Each element in the three-dimensional strategy matrix corresponds to a scenario-physiological-loss combination and is used to store fitting strategy information under that combination. The fitting strategy information includes gain adjustment parameters and noise reduction parameters.

[0016] A4: Assign initial weights to fitting strategies under different scenario-physiology-loss combinations;

[0017] A5: For environmental acoustic feature data, physiological indicator data, and motion data, Kalman filter models are established separately. The Kalman filter prediction results of different data sources are weighted and fused to obtain fused time series data.

[0018] A6: For the questionable parts of the fused time series data, a basic probability distribution function is defined for each fitting data source. The fitting data is mapped to different propositions and assigned corresponding probability values. At the same time, the basic probability distribution functions of different fitting data sources are combined using the synthesis rules of DS evidence theory to obtain high-confidence fused data. The questionable parts are measurement errors and individual differences in physiological indicator data.

[0019] A7: Feature extraction is performed on the high-confidence fusion data to obtain feature vectors of different feature dimensions. The importance weight of each feature dimension is calculated through the attention mechanism, and the final attention fusion fitting data is obtained after weighted fusion.

[0020] Specifically, the constructing of the scene-physiology-loss three-dimensional adaptive model further includes:

[0021] A8: A machine learning algorithm is used to construct an environmental scene recognition sub-model with the environmental acoustic feature data and motion data in the attention fusion fitting data as input. A physiological state monitoring sub-model is constructed with the physiological indicator data and environmental acoustic feature data in the attention fusion fitting data as input.

[0022] A9: By analyzing equipment usage data and regular calibration results, combined with current input data, a loss characteristic analysis sub-model is established;

[0023] A10: Map the output results of the environmental scene recognition sub-model, the physiological status monitoring sub-model, and the loss feature analysis sub-model into a three-dimensional feature space to form a scene-physiology-loss three-dimensional adaptive model; in the three-dimensional feature space, each point represents a specific scene-physiology-loss combination.

[0024] Specifically, the compensation parameter generation module dynamically assigns priorities to the three types of data based on the real-time fused multi-source fitting data and the scene-physiology-loss three-dimensional adaptive model data, and generates initial compensation parameters, including:

[0025] B1: Preprocessing the multi-source fitting data after real-time fusion; the preprocessing includes data normalization, removal of outliers and filling in missing values;

[0026] B2: Based on the scenario-physiology-loss three-dimensional adaptive model, determine the reference range of the gain adjustment parameters and noise reduction parameters in the fitting strategy information corresponding to the current scenario-physiology-loss combination;

[0027] B3: Assign initial priority weights to the ambient acoustic feature data based on the noise type and noise intensity in the ambient acoustic feature data and the output results of the ambient scene recognition sub-model;

[0028] B4: Assign initial priority weights to the physiological indicator data based on the degree of hearing loss, hearing change trend, and output results of the physiological status monitoring sub-model;

[0029] B5: Combine the output of the sub-model for analyzing the motion intensity, motion type, and loss characteristics in the motion data to assign an initial priority weight to the motion data;

[0030] B6: Construct a comprehensive priority evaluation model based on fuzzy logic. Input the initial priority weights of environmental acoustic feature data, physiological indicator data, and motion data into the comprehensive priority evaluation model. Consider the mutual influence between the three types of data and obtain the final dynamic priority weights.

[0031] B7: Based on the final dynamic priority weight, weighted processing is performed on the multi-source fitting data after real-time fusion to obtain weighted fusion data;

[0032] B8: Using the weighted fusion data as input, combined with the reference ranges of the corresponding gain adjustment parameters and noise reduction parameters in the scene-physiology-loss three-dimensional adaptive model, a genetic algorithm is used for optimization search to generate initial compensation parameters; the initial compensation parameters include initial gain parameters and initial noise reduction parameters.

[0033] Specifically, when assigning initial priority weights to the environmental acoustic feature data in B3, the following steps are included:

[0034] B3.1: Establish a mapping table between noise types and priority weights. Different types of noise correspond to different initial priority weight adjustment values.

[0035] B3.2: Dynamically adjust the initial priority weight based on the comparison between the noise intensity and the preset noise threshold;

[0036] When the noise intensity exceeds the preset noise threshold, the priority weight of the environmental acoustic feature data is increased according to the preset proportional coefficient;

[0037] B3.3: Based on the output results of the environmental scene recognition sub-model, if the identified environmental scene is an environmental sound focus scene, the priority weight of the environmental acoustic feature data is increased according to the dynamic proportional coefficient; the environmental sound focus scene refers to a scene that requires attention to environmental sound.

[0038] Specifically, the initial compensation parameters are input into the third-order LSTM acoustic compensation engine configured in the acoustic compensation module, and the acoustic compensation signal is output in combination with the audio signal collected according to the predefined real-time hearing test strategy, including:

[0039] C1: According to a predefined real-time hearing test strategy, audio signals are collected under different scene-physiological-loss combinations; the audio signals include original audio signals and audio signals after initial compensation;

[0040] The predefined real-time hearing test strategy is as follows: every 500 milliseconds, a set of test audio signals of varying frequencies and intensities is input to the conduction device, while simultaneously collecting the processed output audio signal from the conduction device, comparing the difference between the input and output audio signals, and obtaining the current acoustic performance data of the conduction device;

[0041] C2: Preprocess the collected audio signal;

[0042] C3: Input the preprocessed audio signal into a third-order LSTM acoustic compensation engine; the third-order LSTM acoustic compensation engine includes three LSTM layers, each of which is used to extract features of the audio signal at different time scales;

[0043] C4: In the third-order LSTM acoustic compensation engine, the initial compensation parameters are used as input information and fused with the audio signal features to obtain the fused audio signal features;

[0044] C5: The third-order LSTM acoustic compensation engine calculates the preliminary acoustic compensation signal based on the input fused audio signal features through forward propagation.

[0045] C6: performing post-processing on the preliminary result of the acoustic compensation signal; the post-processing includes limiting and smoothing operations;

[0046] C7: Outputs the post-processed acoustic compensation signal.

[0047] Specifically, the parameters of each LSTM layer of the third-order LSTM acoustic compensation engine in C3 are set as follows:

[0048] C3.1: The first LSTM layer is used to extract short-term features of the audio signal. The number of hidden layer nodes is set to a.

[0049] C3.2: The second LSTM layer is used to extract the mid-term features of the audio signal. The number of hidden layer nodes is set to b.

[0050] C3.3: The third layer of LSTM is used to extract the long-term features of the audio signal. The number of hidden layer nodes is set to c and it satisfies ;

[0051] C3.4: Each LSTM layer uses dropout technology to randomly discard a preset number of neurons.

[0052] Specifically, the execution and feedback module converts the acoustic compensation signal into executable instructions for the main control chip of the conduction device, drives the conduction device to perform acoustic compensation, and outputs the gain curve in real time. At the same time, it collects user feedback information on the current acoustic compensation effect, including:

[0053] D1: Convert the acoustic compensation signal into executable instructions for the main control chip of the conduction device according to the instruction format requirements of the main control chip of the conduction device;

[0054] The process of converting the acoustic compensation signal into an executable instruction of the main control chip of the conduction device is as follows: converting the digital acoustic compensation signal into an analog signal through a digital-to-analog conversion chip, and then encoding the analog signal into a digital code that conforms to the instruction format of the main control chip using a communication protocol, and transmitting the digital code to the main control chip;

[0055] D2: Sends the executable instructions of the main control chip to the main control chip of the conduction device, driving the acoustic compensation module of the conduction device to perform acoustic compensation operations;

[0056] D3: During the acoustic compensation process, the output signal of the conduction device is collected in real time, and the gain curve is calculated based on the output signal;

[0057] D4: Displaying the current acoustic compensation effect to the user through the user interface or wireless communication, and collecting user feedback on the acoustic compensation effect; the feedback information includes the user's evaluation of sound clarity, comfort, and volume;

[0058] D5: Quantify user feedback information and convert it into numerical indicators.

[0059] Specifically, the calculation of the gain curve in D3 includes:

[0060] D3.1: Determine the frequency range and number of sampling points of the gain curve; the frequency range covers the audio frequency range that can be processed by the transmission device;

[0061] D3.2: At each sampling point, calculate the amplitude ratio of the conduction device's output signal to the input signal to obtain the gain value at the sound wave frequency corresponding to the current sampling point;

[0062] D3.3: Connect the gain values ​​of all sampling points to form a gain curve, and smooth the gain curve.

[0063] A conduction device, comprising a conduction device body, a data acquisition sensor group, a main control chip, an acoustic compensation module and a user interaction interface;

[0064] The conducting device body is the main structure of the device and is used to install and fix components;

[0065] The data acquisition sensor group is arranged on the main body of the conduction device, and is used to collect environmental acoustic characteristic data of the user's environment, the user's physiological index data and the user's motion data, and transmit the collected data to the main control chip;

[0066] The main control chip receives data transmitted by the data acquisition sensor group and runs the conduction device fitting management system. It dynamically generates compensation parameters based on the real-time fused multi-source fitting data and the scene-physiology-loss three-dimensional adaptive model data, and controls the acoustic compensation module to perform acoustic compensation.

[0067] The acoustic compensation module receives the compensation parameters transmitted by the main control chip, performs compensation processing on the received audio signal, and outputs the compensated audio signal;

[0068] The user interaction interface is arranged on the main body of the conduction device, and is used to display the current acoustic compensation effect, receive user feedback information on the acoustic compensation effect, and transmit the feedback information to the main control chip.

[0069] Compared with the prior art, the present invention has the following beneficial effects:

[0070] 1. The present invention proposes a conduction device fitting management system, and optimizes and improves the architecture, operating steps and processes. The system has the advantages of simple process, low investment and operating costs, and low production work costs.

[0071] 2. The present invention proposes a conduction device fitting management system, which uses a three-dimensional modeling module to fuse multi-source fitting data in real time and combines it with a regional fitting strategy to construct a scene-physiology-loss three-dimensional adaptive model. The compensation parameter generation module dynamically assigns priorities to three types of data and generates initial compensation parameters based on this model and real-time fusion data. This collaborative mechanism enables the conduction device to accurately adapt to different scenarios, physiological states, and equipment loss conditions, breaking through the limitations of traditional fixed compensation modes, realizing personalized acoustic compensation, and effectively improving the accuracy and adaptability of acoustic compensation, ensuring that users can obtain a hearing experience that better suits their needs in different environments.

[0072] 3. The present invention proposes a conduction device fitting management system that inputs initial compensation parameters into a third-order LSTM acoustic compensation engine. This engine, combined with audio signals collected using a real-time hearing test strategy, outputs acoustic compensation signals. The execution and feedback module then converts these signals into executable instructions to drive the conduction device for compensation. The engine also outputs gain curves and collects user feedback in real time. This process not only enables efficient generation and accurate execution of acoustic compensation signals, but also enables timely understanding of user perceptions of the compensation effect through a real-time feedback mechanism. This contributes to the continuous optimization of compensation parameters and algorithms, further improving the acoustic compensation effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] Figure 1 This is an architecture diagram of a conduction device fitting management system of the present invention;

[0074] Figure 2 A flow chart for constructing a scenario-physiology-loss three-dimensional adaptive model for a conduction device fitting management system according to the present invention;

[0075] Figure 3 This is a flow chart for implementing initial compensation parameters of a conduction device fitting management system of the present invention. DETAILED DESCRIPTION

[0076] Example 1

[0077] See also Figure 1 An embodiment of the present invention provides a conductive device fitting management system applicable to the hearing aid application field. In the hearing aid application field, loss refers to hearing loss, and conductive device refers to hearing aid, including: a data acquisition module, a three-dimensional modeling module, a compensation parameter generation module, an acoustic compensation module, and an execution and feedback module;

[0078] The three-dimensional modeling module performs real-time fusion of the multi-source fitting data collected by the data acquisition module, and combines it with a pre-defined regional fitting strategy to build a scene-physiology-loss three-dimensional adaptive model; the multi-source fitting data includes environmental acoustic feature data, physiological indicator data, and motion data;

[0079] Environmental acoustic characteristic data: A high-precision microphone array is used to collect ambient sound signals in multiple directions. The microphone array can obtain information such as sound intensity, frequency distribution, and phase. Through beamforming technology, it can determine the direction of the main noise source and calculate the sound pressure level at different frequencies. For example, in an office scene, there are noise sources such as computer fans and printers. The microphone array can accurately measure the sound pressure level and frequency characteristics of these noises.

[0080] Physiological indicator data: Wearable biosensors such as heart rate monitors, oximeters, and galvanic skin response sensors collect real-time data on the user's heart rate, blood oxygen saturation, and galvanic skin response. These sensors can accurately measure changes in physiological parameters, reflecting the user's physiological state in different scenarios. For example, when a user is stressed or fatigued, their heart rate may increase and their galvanic skin response may be enhanced.

[0081] Motion data: Utilizes inertial sensors such as accelerometers and gyroscopes to collect information such as the user's motion acceleration and angular velocity. Calculate parameters such as the user's motion speed and displacement through an integration algorithm to determine the user's motion state, such as stillness, walking, or running.

[0082] The scenario-physiology-loss three-dimensional adaptive model can dynamically reflect the optimal fitting parameters under different listening scenarios, user physiological states and hearing loss conditions.

[0083] It should be noted that different areas may have different acoustic environment characteristics and fitting requirements. For example, in office scenarios, more emphasis is placed on compensating for speech clarity, while in noisy outdoor environments, the focus is on improving the overall signal-to-noise ratio. Therefore, different fitting strategies need to be defined according to different situations.

[0084] The compensation parameter generation module dynamically assigns priorities to the three types of data based on the real-time fused multi-source fitting data and the scene-physiology-loss three-dimensional adaptive model data, and generates initial compensation parameters;

[0085] Wherein, the priority allocation is dynamically adjusted based on the degree of influence of the data on the hearing compensation effect;

[0086] The initial compensation parameters include but are not limited to gain, compression ratio, and noise reduction threshold.

[0087] For example, in different scenarios and user states, the degree of influence of various types of data on the compensation parameters is different. For example, in a noisy environment, the priority of environmental acoustic feature data is higher; when the user is in a nervous state, the priority of physiological indicator data needs to be increased.

[0088] Inputting the initial compensation parameters into a third-order LSTM acoustic compensation engine configured in the acoustic compensation module, combining the audio signal collected according to a predefined real-time hearing test strategy, outputting an acoustic compensation signal, and transmitting it to the execution and feedback module;

[0089] The execution and feedback module converts the acoustic compensation signal into executable instructions for the main control chip of the conduction device, drives the conduction device to perform acoustic compensation, and outputs the gain curve in real time. At the same time, it collects user feedback information on the current acoustic compensation effect.

[0090] It should be noted that the execution and feedback module converts the acoustic compensation signal into actual executable instructions to drive the hearing aid to work, and transmits feedback information back to the data acquisition module, so that the system can continuously adjust and optimize the subsequent fitting process according to the user's actual experience, forming an adaptive closed-loop system to continuously improve the acoustic compensation effect of the hearing aid.

[0091] The data acquisition module includes: an acoustic sensor unit, a physiological indicator sensor unit, and a motion sensor unit;

[0092] An acoustic sensor unit, used for collecting environmental acoustic characteristic data;

[0093] A physiological indicator sensor unit, used for collecting physiological indicator data;

[0094] The motion sensor unit is used to collect motion data.

[0095] The three-dimensional modeling module includes: a data fusion unit and a model building unit;

[0096] Data fusion unit, used for time synchronization and normalization of multi-source fitting data;

[0097] The model building unit is used to combine the regional testing and fitting strategy to build a three-dimensional adaptive model of scenario-physiology-loss.

[0098] The compensation parameter generation module includes: a priority allocation unit and a parameter generation unit;

[0099] A priority allocation unit, used to dynamically allocate the priorities of environmental acoustic feature data, physiological indicator data, and motion data;

[0100] The parameter generating unit is used to generate initial compensation parameters according to the priority.

[0101] The acoustic compensation module includes: an engine configuration unit and a signal processing unit;

[0102] The engine configuration unit is used to configure the structure and parameters of the third-order LSTM acoustic compensation engine.

[0103] The signal processing unit is used to output an acoustic compensation signal in combination with the audio signal collected by the real-time hearing test strategy.

[0104] Example 2

[0105] See also Figure 2 and Figure 3 In this embodiment, the three-dimensional modeling module performs real-time fusion of the multi-source fitting data collected by the data acquisition module, and combines it with a pre-defined regional fitting strategy to build a scene-physiology-loss three-dimensional adaptive model, including:

[0106] A1: Collect multi-source fitting data; the multi-source fitting data includes environmental acoustic feature data, physiological index data, and motion data;

[0107] Furthermore, the method for collecting the environmental acoustic feature data is: using multiple microphone arrays of different types, arranged at different positions of the conduction device, to collect environmental sound signals in different directions and frequency bands, converting the time domain sound signal into a frequency domain signal through Fourier transform, and extracting environmental acoustic feature data including environmental noise intensity, main noise frequency components, and reverberation time. Among them, Fourier transform is the existing technical content in this field and is not the creative solution of this application, so it will not be elaborated here.

[0108] Furthermore, the physiological indicator data includes ear canal shape data, tympanic membrane thickness data, and auditory nerve response data; the ear canal shape data is obtained through three-dimensional laser scanning technology, the tympanic membrane thickness data is collected using ultrasonic measurement technology, and the auditory nerve response data is recorded with the help of electroencephalogram signal acquisition equipment under specific audio stimulation.

[0109] Furthermore, the motion data uses a nine-axis inertial sensor to collect the acceleration, angular velocity, and magnetic field strength data of the wearer of the conduction device in real time, and the sensor data is fused and processed through a complementary filtering algorithm to extract the wearer's motion data such as motion state, motion amplitude, and motion direction.

[0110] A2: defining a regionalized fitting strategy; the regionalized fitting strategy includes defining an environmental scene category set, a physiological state category set, and a loss feature set;

[0111] Define a set of environmental scene categories, such as quiet, noisy, and music scenes;

[0112] Define a set of physiological state categories, such as relaxation, tension, and exercise;

[0113] Define a set of loss characteristics, such as frequency response curves and sensitivity thresholds.

[0114] A3: Establish a three-dimensional strategy matrix based on the environmental scenario category, physiological state category, and loss characteristics. Each element in the three-dimensional strategy matrix corresponds to a scenario-physiological-loss combination and is used to store fitting strategy information under that combination. The fitting strategy information includes gain adjustment parameters and noise reduction parameters.

[0115] Furthermore, the three-dimensional strategy matrix is ​​stored in a hash table data structure, with the scenario-physiology-loss combination as the hash key and the fitting strategy information as the hash value; when performing data query and call, the corresponding fitting strategy information is quickly located through the hash function, thereby improving the operating efficiency of the system.

[0116] A4: Assign initial weights to fitting strategies under different scenario-physiology-loss combinations;

[0117] A5: For environmental acoustic feature data, physiological indicator data, and motion data, Kalman filter models are established separately. The Kalman filter prediction results of different data sources are weighted and fused to obtain fused time series data.

[0118] For example, taking environmental acoustic characteristic data as an example, parameters such as sound pressure level and frequency are used as state variables. According to historical data and current measurement values, the Kalman filter algorithm is used for state estimation and prediction. By continuously updating the state estimation value and covariance matrix, the prediction accuracy of the data change trend is improved. Among them, the Kalman filter algorithm is the existing technology content in this field and is not the creative solution of this application, so it will not be elaborated here.

[0119] A6: For the questionable parts of the fused time series data, a basic probability distribution function is defined for each fitting data source. The fitting data is mapped to different propositions and assigned corresponding probability values. At the same time, the basic probability distribution functions of different fitting data sources are combined using the synthesis rules of DS evidence theory to obtain high-confidence fused data. The questionable parts are measurement errors and individual differences in physiological indicator data.

[0120] For example, for heart rate data, map it to propositions such as calm, nervous, and tired, and assign corresponding probabilities.

[0121] A7: Feature extraction is performed on the high-confidence fusion data to obtain feature vectors of different feature dimensions. The importance weight of each feature dimension is calculated through the attention mechanism. After weighted fusion, the final attention fusion fitting data is obtained;

[0122] A8: A machine learning algorithm is used to construct an environmental scene recognition sub-model with the environmental acoustic feature data and motion data in the attention fusion fitting data as input. A physiological state monitoring sub-model is constructed with the physiological indicator data and environmental acoustic feature data in the attention fusion fitting data as input.

[0123] A9: By analyzing equipment usage data and regular calibration results, combined with current input data, a loss characteristic analysis sub-model is established;

[0124] Furthermore, the environmental scene recognition sub-model automatically identifies the current environmental scene category based on the characteristics of the input fitting data and outputs it; the physiological state monitoring sub-model monitors the user's physiological state in real time and classifies it into the corresponding physiological state category; the loss feature analysis sub-model estimates the loss characteristics of the equipment, such as the high-frequency loss degree, low-frequency loss degree, etc., and outputs it.

[0125] A10: Map the output results of the environmental scene recognition sub-model, the physiological status monitoring sub-model, and the loss feature analysis sub-model into a three-dimensional feature space to form a scene-physiology-loss three-dimensional adaptive model; in the three-dimensional feature space, each point represents a specific scene-physiology-loss combination.

[0126] The compensation parameter generation module dynamically assigns priorities to the three types of data based on the real-time fused multi-source fitting data and the scene-physiology-loss three-dimensional adaptive model data, and generates initial compensation parameters, including:

[0127] B1: Preprocessing the multi-source fitting data after real-time fusion; the preprocessing includes data normalization, removal of outliers and filling in missing values;

[0128] B2: Based on the scenario-physiology-loss three-dimensional adaptive model, determine the reference range of the gain adjustment parameters and noise reduction parameters in the fitting strategy information corresponding to the current scenario-physiology-loss combination;

[0129] B3: Assigning initial priority weights to the ambient acoustic feature data based on the noise type and noise intensity in the ambient acoustic feature data and the output of the ambient scene recognition sub-model; for example, when the noise intensity exceeds a preset noise threshold and the ambient scene is identified as a noisy public place, the priority weight of the ambient acoustic feature data is increased;

[0130] Furthermore, the noise threshold setting needs to be combined with the specific application scenario, device performance, and human perception characteristics. It is usually determined by using a baseline reference value + dynamic scenario adjustment method, including:

[0131] (1) Benchmark threshold setting based on acoustic standards

[0132] Based on internationally accepted acoustic standards and combining people's subjective perception of different noise intensities, the unit is decibel (dB(A). A-weighting is more in line with human ear perception:

[0133] Quiet environment benchmark: For scenarios such as residential and office environments, the preset basic threshold is usually 40-50dB(A). When the noise intensity is less than or equal to this preset basic threshold, it is judged as low noise and the environmental acoustic data has a lower priority.

[0134] Noisy environment benchmark: For scenarios such as streets and shopping malls, the preset benchmark threshold is usually 60-70dB(A). When the noise intensity exceeds this preset benchmark threshold, it is judged as medium-to-high noise, and the priority of environmental acoustic data needs to be increased;

[0135] (2) Dynamic adjustment based on scene type

[0136] Based on the baseline threshold, the threshold is further refined according to the scene category output by the environmental scene recognition sub-model:

[0137] In quiet scenarios, such as libraries and bedrooms, the threshold is lowered to 30-40dB(A). Even relatively low noise levels, such as a whisper at 35dB(A), can affect the user experience, so acoustic data priority should be increased.

[0138] Noisy public places, such as subways and concerts: The threshold is raised to 70-85dB(A), as the background noise level in these scenarios is high.

[0139] Dynamic scenes, such as driving or walking: Set a threshold range, such as 50-75dB(A), and adjust it in real time based on noise fluctuations in the scene. For example, when the vehicle speed increases, the threshold is increased by 10-15dB(A).

[0140] B4: Assign initial priority weights to the physiological indicator data based on the degree of hearing loss, hearing change trends, and the output of the physiological status monitoring sub-model. For example, if the hearing loss is severe and the physiological status monitoring indicates fatigue, appropriately increase the priority weight of the physiological indicator data.

[0141] Furthermore, initial priority weights are assigned to the physiological indicator data, including:

[0142] B4.1: Establish a functional relationship between the degree of hearing loss and the priority weight. The more severe the hearing loss, the higher the corresponding initial priority weight.

[0143] B4.2: Based on the hearing change trend, if hearing declines within a short period of time, increase the priority weight of physiological indicator data;

[0144] B4.3: Based on the output of the physiological status monitoring sub-model, if it is detected that the user is in an adverse physiological state such as tension or anxiety that affects hearing perception, increase the priority weight of the physiological indicator data.

[0145] B5: Assign an initial priority weight to the motion data based on the motion intensity, motion type, and the output of the wear and tear feature analysis sub-model. For example, if the motion intensity is high and the wear and tear feature analysis indicates increased device wear, increase the priority weight of the motion data.

[0146] Furthermore, initial priority weights are assigned to the motion data, including:

[0147] B5.1: Establish a corresponding relationship between exercise intensity and priority weight. The greater the exercise intensity, the higher the corresponding initial priority weight.

[0148] B5.2: Based on the type of exercise, if the exercise type involves severe vibration, such as running or jumping, the priority weight of the exercise data will be increased;

[0149] B5.3: Combined with the output results of the loss feature analysis sub-model, if the device loss level is high and related to motion, further increase the priority weight of the motion data.

[0150] B6: Construct a comprehensive priority evaluation model based on fuzzy logic. Input the initial priority weights of environmental acoustic feature data, physiological indicator data, and motion data into the comprehensive priority evaluation model. Consider the mutual influence between the three types of data and obtain the final dynamic priority weights.

[0151] Furthermore, when constructing a comprehensive priority evaluation model based on fuzzy logic, it includes:

[0152] B6.1: Determine the initial priority weights of the fuzzy input variables, namely, environmental acoustic characteristic data, physiological indicator data, and motion data;

[0153] B6.2: Define fuzzy sets. Define different fuzzy sets for each fuzzy input variable, such as low priority, medium priority, and high priority.

[0154] B6.3: Establish a fuzzy rule base and formulate fuzzy rules based on actual application scenarios and expert experience to describe the relationship between different fuzzy input variable combinations and the final priority weights;

[0155] B6.4: Using fuzzy reasoning methods to perform reasoning based on fuzzy input variables and a fuzzy rule base to obtain fuzzy output results. Fuzzy reasoning methods are prior art in this field and do not constitute the inventive solution of this application, so they are not described in detail here.

[0156] B6.5: Defuzzify the fuzzy output results to obtain the final dynamic priority weight.

[0157] B7: Based on the final dynamic priority weight, weighted processing is performed on the multi-source fitting data after real-time fusion to obtain weighted fusion data;

[0158] B8: Using the weighted fusion data as input, combined with the reference ranges of the corresponding gain adjustment parameters and noise reduction parameters in the scene-physiology-loss three-dimensional adaptive model, a genetic algorithm is used for optimization search to generate initial compensation parameters; the initial compensation parameters include initial gain parameters and initial noise reduction parameters.

[0159] Furthermore, a genetic algorithm is used for optimization search to generate initial compensation parameters, including:

[0160] B8.1: Determine the encoding method of the genetic algorithm and encode the initial gain parameters and initial noise reduction parameters into chromosomes;

[0161] B8.2: Define a fitness function that comprehensively considers the degree to which the compensation parameters match the ambient acoustic characteristic data, physiological indicator data, and motion data, as well as their proximity to the reference range in the scene-physiology-loss three-dimensional adaptive model.

[0162] B8.3: Set the parameters of the genetic algorithm, including population size, crossover probability, and mutation probability. The genetic algorithm is prior art in this field and does not constitute an inventive solution of this application, so it will not be described in detail here.

[0163] B8.4: Perform the population initialization operation to randomly generate a predetermined number of chromosomes as the initial population;

[0164] B8.5: Perform selection, crossover, and mutation operations. Individuals in the population are selected based on the fitness function, and new individuals are generated through crossover and mutation. The selection, crossover, and mutation operations are known in the art and do not constitute the inventive solution of this application. These operations are not described in detail here.

[0165] B8.6: Repeat the selection, crossover, and mutation operations until the preset termination conditions are met, such as reaching the preset maximum number of iterations or the fitness function value reaching the preset threshold, and output the optimal chromosome, that is, the initial compensation parameters.

[0166] When allocating initial priority weights to the environmental acoustic feature data in B3, the following steps are included:

[0167] B3.1: Establish a mapping table between noise types and priority weights. Different types of noise correspond to different initial priority weight adjustment values. For example, sudden high-frequency noise corresponds to a larger initial priority weight adjustment value.

[0168] B3.2: Dynamically adjust the initial priority weight based on the comparison between the noise intensity and the preset noise threshold;

[0169] When the noise intensity exceeds the preset noise threshold, the priority weight of the environmental acoustic feature data is increased according to the preset proportional coefficient;

[0170] Furthermore, the preset proportional coefficient is used to quantify the increase in the priority weight of the environmental acoustic feature data when the noise intensity exceeds the threshold. Its setting needs to balance response sensitivity and system stability. It is usually determined comprehensively based on the degree of noise exceeding the standard or the characteristics of the scene. The specific method is as follows:

[0171] (1) Step-by-step proportional coefficient based on the noise exceeding the standard:

[0172] The noise intensity is divided into intervals according to the difference between the preset threshold and the noise intensity is divided into intervals. A fixed proportional coefficient is assigned to each interval to avoid over-adjustment or under-adjustment caused by a single coefficient. For mild exceedance (ΔdB≤10dB), the proportional coefficient is set to 1.2-1.5; for moderate exceedance (10dB<ΔdB≤20dB), the proportional coefficient is set to 1.5-2.0; for severe exceedance (ΔdB>20dB), the proportional coefficient is set to 2.0-3.0;

[0173] (2) Dynamic scaling coefficients based on scene categories

[0174] Different scenarios have different tolerances for noise, and exclusive proportional coefficients need to be matched for specific scenarios to avoid unreasonable adjustments caused by unified adjustments across scenarios. This includes: in quiet scenarios, even if the standard is slightly exceeded, the proportional coefficient is set to 1.5-2.0. This is because quiet environments are more sensitive to noise and the priority of acoustic data needs to be increased; in noisy public places, when the standard is moderately exceeded, the proportional coefficient is set to 1.2-1.5 because the background noise of the scene is high; in dynamic scenarios, the coefficient is adjusted according to the intensity of activities in the scene. For example, when the vehicle speed is greater than 60km / h, the proportional coefficient is reduced by 10%-20% to avoid frequent triggering of weight adjustments due to noise fluctuations caused by vehicle bumps.

[0175] B3.3: Based on the output results of the environmental scene recognition sub-model, if the identified environmental scene is an environmental sound focus scene, the priority weight of the environmental acoustic feature data is increased according to the dynamic proportional coefficient; the environmental sound focus scene refers to a scene that requires high attention to environmental sounds, including but not limited to the examination site.

[0176] Inputting the initial compensation parameters into the third-order LSTM acoustic compensation engine configured in the acoustic compensation module, combining the hearing audio signal collected according to the predefined real-time hearing test strategy, and outputting the acoustic compensation signal, including:

[0177] C1: According to a predefined real-time hearing test strategy, audio signals are collected under different scene-physiological-loss combinations; the audio signals include original audio signals and audio signals after initial compensation;

[0178] The predefined real-time hearing test strategy is as follows: every 500 milliseconds, a set of test audio signals of varying frequencies and intensities is input to the conduction device, while simultaneously collecting the processed output audio signal from the conduction device, comparing the difference between the input and output audio signals, and obtaining the current acoustic performance data of the conduction device;

[0179] C2: Preprocess the collected audio signal; preprocessing includes pre-emphasis, framing, windowing and other operations to improve the quality of the audio signal and the accuracy of feature extraction;

[0180] C3: Input the preprocessed audio signal into a third-order LSTM acoustic compensation engine; the third-order LSTM acoustic compensation engine includes three LSTM layers, each of which is used to extract features of the audio signal at different time scales;

[0181] Furthermore, the internal structure of the third-order LSTM acoustic compensation engine includes three layers of long short-term memory network layers, and each LSTM layer contains 256 memory units.

[0182] C4: In the third-order LSTM acoustic compensation engine, the initial compensation parameters are used as input information and fused with the audio signal features to obtain the fused audio signal features;

[0183] Furthermore, when the initial compensation parameters are used as input information and fused with the audio signal features, the following steps are included:

[0184] C4.1: Normalize the initial compensation parameters to match the range of audio signal characteristics;

[0185] C4.2: Concatenate the normalized initial compensation parameters with the output features of each LSTM layer to form a new fused feature vector;

[0186] C4.3: During the splicing process, different weights are assigned to the initial compensation parameters and audio signal features to adjust their importance in the fused feature vector and obtain the fused audio signal features.

[0187] C5: The third-order LSTM acoustic compensation engine calculates the preliminary acoustic compensation signal based on the input fused audio signal features through forward propagation.

[0188] C6: performing post-processing on the preliminary result of the acoustic compensation signal; the post-processing includes limiting and smoothing operations to avoid distortion or mutation of the acoustic compensation signal;

[0189] C7: Outputs the post-processed acoustic compensation signal.

[0190] The parameters of each LSTM layer of the third-order LSTM acoustic compensation engine in C3 are set as follows:

[0191] C3.1: The first LSTM layer is used to extract short-term features of the audio signal. The number of hidden layer nodes is set to a to capture local changes in the audio signal.

[0192] C3.2: The second LSTM layer is used to extract the medium-term features of the audio signal. The number of hidden layer nodes is set to b, and both short-term and long-term features are considered.

[0193] C3.3: The third layer of LSTM is used to extract the long-term features of the audio signal. The number of hidden layer nodes is set to c and it satisfies , to capture the overall trend of the audio signal;

[0194] C3.4: Each LSTM layer uses dropout technology to randomly discard a preset number of neurons to prevent overfitting.

[0195] The execution and feedback module converts the acoustic compensation signal into executable instructions for the main control chip of the conduction device, drives the conduction device to perform acoustic compensation, and outputs the gain curve in real time. At the same time, it collects user feedback on the current acoustic compensation effect, including:

[0196] D1: Convert the acoustic compensation signal into executable instructions for the main control chip of the conduction device according to the instruction format requirements of the main control chip of the conduction device;

[0197] The process of converting the acoustic compensation signal into an executable instruction of the main control chip of the conduction device is as follows: converting the digital acoustic compensation signal into an analog signal through a digital-to-analog conversion chip, and then encoding the analog signal into a digital code that conforms to the instruction format of the main control chip using a communication protocol, and transmitting the digital code to the main control chip;

[0198] D2: Sends the executable instructions of the main control chip to the main control chip of the conduction device, driving the acoustic compensation module of the conduction device to perform acoustic compensation operations;

[0199] D3: During the acoustic compensation process, the output signal of the conduction device is collected in real time, and the gain curve is calculated based on the output signal. The gain curve reflects the gain change at different frequencies;

[0200] D4: Displaying the current acoustic compensation effect to the user through the user interface or wireless communication, and collecting user feedback on the acoustic compensation effect; the feedback information includes the user's evaluation of sound clarity, comfort, and volume;

[0201] Furthermore, collecting user feedback on the acoustic compensation effect includes: setting up star ratings and text evaluation input boxes in the system, and equipping it with a voice feedback function. Users can provide feedback on their satisfaction with the current acoustic compensation effect and improvement suggestions by clicking on star ratings, entering text or voice descriptions.

[0202] D5: Quantify user feedback information and convert it into numerical indicators that can be used for subsequent analysis and optimization.

[0203] Furthermore, the user feedback information is quantified, including:

[0204] D5.1: Assign a numerical score to each feedback rating option, e.g., a five-star rating corresponds to 5 points and a one-star rating corresponds to 1 point;

[0205] D5.2: Calculate the average or weighted average score of each feedback indicator based on the evaluation options selected by the user;

[0206] D5.3: Store the quantified feedback information in a database for subsequent data analysis and system optimization.

[0207] The calculation of the gain curve in D3 includes:

[0208] D3.1: Determine the frequency range and number of sampling points of the gain curve; the frequency range covers the audio frequency range that can be processed by the transmission device;

[0209] D3.2: At each sampling point, calculate the amplitude ratio of the conduction device's output signal to the input signal to obtain the gain value at the sound wave frequency corresponding to the current sampling point;

[0210] Furthermore, in acoustic or signal processing scenarios, performance analysis of a conduction device typically needs to cover a preset frequency range, such as 20Hz-20kHz for human-audible sound. During analysis, multiple sampling points are selected at preset intervals, with each sampling point corresponding to a specific frequency value, such as 100Hz, 500Hz, or 1kHz.

[0211] D3.3: Connect the gain values ​​of all sampling points to form a gain curve, and smooth the gain curve to reduce noise interference.

[0212] Example 3

[0213] A conduction device, comprising a conduction device body, a data acquisition sensor group, a main control chip, an acoustic compensation module and a user interaction interface;

[0214] The conducting device body is the main structure of the device and is used to install and fix components;

[0215] The data acquisition sensor group is arranged on the main body of the conduction device, and is used to collect environmental acoustic characteristic data of the user's environment, the user's physiological index data and the user's motion data, and transmit the collected data to the main control chip;

[0216] The main control chip receives data transmitted by the data acquisition sensor group and runs the conduction device fitting management system. It dynamically generates compensation parameters based on the real-time fused multi-source fitting data and the scene-physiology-loss three-dimensional adaptive model data, and controls the acoustic compensation module to perform acoustic compensation.

[0217] The acoustic compensation module receives the compensation parameters transmitted by the main control chip, performs compensation processing on the received audio signal, and outputs the compensated audio signal;

[0218] The user interaction interface is arranged on the main body of the conduction device, and is used to display the current acoustic compensation effect, receive user feedback information on the acoustic compensation effect, and transmit the feedback information to the main control chip.

[0219] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific embodiments. The above-mentioned specific embodiments are merely illustrative and not restrictive. Under the guidance of the present invention, ordinary technicians in this field can also change, modify, replace and modify the above-mentioned embodiments without departing from the purpose and scope of protection of the present invention. These are all protected by the present invention.

[0220] If the technical solution disclosed herein involves personal information, the product using the technical solution disclosed herein has clearly informed the individual of the personal information processing rules and obtained the individual's voluntary consent before processing the personal information. If the technical solution disclosed herein involves sensitive personal information, the product using the technical solution disclosed herein has obtained the individual's separate consent before processing the sensitive personal information and at the same time meets the "explicit consent" requirement. For example, on personal information collection devices such as cameras, a clear and prominent sign is set to inform that the individual has entered the personal information collection scope and that personal information will be collected. If the individual voluntarily enters the collection scope, it is deemed that they agree to the collection of their personal information; or on the personal information processing device, when the personal information processing rules are notified by obvious signs / information, the individual's authorization is obtained through pop-up information or by asking the individual to upload their personal information. The personal information processing rules may include information such as the personal information processor, the purpose of personal information processing, the processing method, and the type of personal information processed.

Claims

1. A conduction device fitting management system, characterized in that: include: Data acquisition module, 3D modeling module, compensation parameter generation module, acoustic compensation module, execution and feedback module; The three-dimensional modeling module performs real-time fusion of the multi-source fitting data collected by the data acquisition module and constructs a scene-physiology-loss three-dimensional adaptive model in combination with a pre-defined regional fitting strategy; The compensation parameter generation module dynamically assigns priorities to the three types of data based on the real-time fused multi-source fitting data and the scene-physiology-loss three-dimensional adaptive model data, and generates initial compensation parameters; Inputting the initial compensation parameters into a third-order LSTM acoustic compensation engine configured in the acoustic compensation module, combining the audio signal collected according to a predefined real-time hearing test strategy, outputting an acoustic compensation signal, and transmitting it to the execution and feedback module; The execution and feedback module converts the acoustic compensation signal into executable instructions for the main control chip of the conduction device, drives the conduction device to perform acoustic compensation, and outputs the gain curve in real time. At the same time, it collects user feedback information on the current acoustic compensation effect.

2. A conduction device fitting management system according to claim 1, characterized in that: The three-dimensional modeling module performs real-time fusion of the multi-source fitting data collected by the data acquisition module and constructs a scenario-physiology-loss three-dimensional adaptive model in combination with a pre-defined regional fitting strategy, including: A1: Collect multi-source fitting data; the multi-source fitting data includes environmental acoustic feature data, physiological index data, and motion data; A2: defining a regionalized fitting strategy; the regionalized fitting strategy includes defining an environmental scene category set, a physiological state category set, and a loss feature set; A3: Establish a three-dimensional strategy matrix based on the environmental scenario category, physiological state category, and loss characteristics. Each element in the three-dimensional strategy matrix corresponds to a scenario-physiological-loss combination and is used to store fitting strategy information under that combination. The fitting strategy information includes gain adjustment parameters and noise reduction parameters. A4: Assign initial weights to fitting strategies under different scenario-physiology-loss combinations; A5: For environmental acoustic feature data, physiological indicator data, and motion data, Kalman filter models are established separately. The Kalman filter prediction results of different data sources are weighted and fused to obtain fused time series data. A6: For the questionable parts of the fused time series data, a basic probability distribution function is defined for each fitting data source. The fitting data is mapped to different propositions and assigned corresponding probability values. At the same time, the basic probability distribution functions of different fitting data sources are combined using the synthesis rules of DS evidence theory to obtain high-confidence fused data. The questionable parts are measurement errors and individual differences in physiological indicator data. A7: Feature extraction is performed on the high-confidence fusion data to obtain feature vectors of different feature dimensions. The importance weight of each feature dimension is calculated through the attention mechanism, and the final attention fusion fitting data is obtained after weighted fusion.

3. A conduction device fitting management system according to claim 2, characterized in that: The constructing of the scene-physiology-loss three-dimensional adaptive model further includes: A8: A machine learning algorithm is used to construct an environmental scene recognition sub-model with the environmental acoustic feature data and motion data in the attention fusion fitting data as input. A physiological state monitoring sub-model is constructed with the physiological indicator data and environmental acoustic feature data in the attention fusion fitting data as input. A9: By analyzing equipment usage data and regular calibration results, combined with current input data, a loss characteristic analysis sub-model is established; A10: Map the output results of the environmental scene recognition sub-model, the physiological status monitoring sub-model, and the loss feature analysis sub-model into a three-dimensional feature space to form a scene-physiology-loss three-dimensional adaptive model; in the three-dimensional feature space, each point represents a specific scene-physiology-loss combination.

4. A conduction device fitting management system according to claim 3, characterized in that: The compensation parameter generation module dynamically assigns priorities to the three types of data based on the real-time fused multi-source fitting data and the scene-physiology-loss three-dimensional adaptive model data, and generates initial compensation parameters, including: B1: Preprocessing the multi-source fitting data after real-time fusion; the preprocessing includes data normalization, removal of outliers and filling in missing values; B2: Based on the scenario-physiology-loss three-dimensional adaptive model, determine the reference range of the gain adjustment parameters and noise reduction parameters in the fitting strategy information corresponding to the current scenario-physiology-loss combination; B3: Assign initial priority weights to the ambient acoustic feature data based on the noise type and noise intensity in the ambient acoustic feature data and the output results of the ambient scene recognition sub-model; B4: Assign initial priority weights to the physiological indicator data based on the degree of hearing loss, hearing change trend, and output results of the physiological status monitoring sub-model; B5: Combine the output of the sub-model for analyzing the motion intensity, motion type, and loss characteristics in the motion data to assign an initial priority weight to the motion data; B6: Construct a comprehensive priority evaluation model based on fuzzy logic. Input the initial priority weights of environmental acoustic feature data, physiological indicator data, and motion data into the comprehensive priority evaluation model. Consider the mutual influence between the three types of data and obtain the final dynamic priority weights. B7: Based on the final dynamic priority weight, weighted processing is performed on the multi-source fitting data after real-time fusion to obtain weighted fusion data; B8: Using the weighted fusion data as input, combined with the reference ranges of the corresponding gain adjustment parameters and noise reduction parameters in the scene-physiology-loss three-dimensional adaptive model, a genetic algorithm is used for optimization search to generate initial compensation parameters; the initial compensation parameters include initial gain parameters and initial noise reduction parameters.

5. A conduction device fitting management system according to claim 4, characterized in that: The assigning of initial priority weights to the environmental acoustic feature data includes: B3.1: Establish a mapping table between noise types and priority weights. Different types of noise correspond to different initial priority weight adjustment values. B3.2: Dynamically adjust the initial priority weight based on the comparison between the noise intensity and the preset noise threshold; When the noise intensity exceeds the preset noise threshold, the priority weight of the environmental acoustic feature data is increased according to the preset proportional coefficient; B3.3: Based on the output results of the environmental scene recognition sub-model, if the identified environmental scene is an environmental sound focus scene, the priority weight of the environmental acoustic feature data is increased according to the dynamic proportional coefficient; the environmental sound focus scene refers to a scene that requires attention to environmental sound.

6. A conduction device fitting management system according to claim 5, characterized in that: Inputting the initial compensation parameters into a third-order LSTM acoustic compensation engine configured in the acoustic compensation module, combining the audio signal collected according to a predefined real-time hearing test strategy, and outputting an acoustic compensation signal, including: C1: According to a predefined real-time hearing test strategy, audio signals are collected under different scene-physiological-loss combinations; the audio signals include original audio signals and audio signals after initial compensation; The predefined real-time hearing test strategy is as follows: every 500 milliseconds, a set of test audio signals of varying frequencies and intensities is input to the conduction device, while simultaneously collecting the processed output audio signal from the conduction device, comparing the difference between the input and output audio signals, and obtaining the current acoustic performance data of the conduction device; C2: Preprocess the collected audio signal; C3: Input the preprocessed audio signal into a third-order LSTM acoustic compensation engine; the third-order LSTM acoustic compensation engine includes three LSTM layers, each of which is used to extract features of the audio signal at different time scales; C4: In the third-order LSTM acoustic compensation engine, the initial compensation parameters are used as input information and fused with the audio signal features to obtain the fused audio signal features; C5: The third-order LSTM acoustic compensation engine calculates the preliminary acoustic compensation signal based on the input fused audio signal features through forward propagation. C6: performing post-processing on the preliminary result of the acoustic compensation signal; the post-processing includes limiting and smoothing operations; C7: Outputs the post-processed acoustic compensation signal.

7. A conduction device fitting management system according to claim 6, characterized in that: The parameters of each LSTM layer of the third-order LSTM acoustic compensation engine are set as follows: C3.1: The first LSTM layer is used to extract short-term features of the audio signal. The number of hidden layer nodes is set to a. C3.2: The second LSTM layer is used to extract the mid-term features of the audio signal. The number of hidden layer nodes is set to b. C3.3: The third layer of LSTM is used to extract the long-term features of the audio signal. The number of hidden layer nodes is set to c and it satisfies ; C3.4: Each LSTM layer uses dropout technology to randomly discard a preset number of neurons.

8. A conduction device fitting management system according to claim 7, characterized in that: The execution and feedback module converts the acoustic compensation signal into executable instructions for the main control chip of the conduction device, drives the conduction device to perform acoustic compensation, and outputs the gain curve in real time. At the same time, it collects user feedback on the current acoustic compensation effect, including: D1: Convert the acoustic compensation signal into executable instructions for the main control chip of the conduction device according to the instruction format requirements of the main control chip of the conduction device; The process of converting the acoustic compensation signal into an executable instruction of the main control chip of the conduction device is as follows: converting the digital acoustic compensation signal into an analog signal through a digital-to-analog conversion chip, and then encoding the analog signal into a digital code that conforms to the instruction format of the main control chip using a communication protocol, and transmitting the digital code to the main control chip; D2: Sends the executable instructions of the main control chip to the main control chip of the conduction device, driving the acoustic compensation module of the conduction device to perform acoustic compensation operations; D3: During the acoustic compensation process, the output signal of the conduction device is collected in real time, and the gain curve is calculated based on the output signal; D4: Displaying the current acoustic compensation effect to the user through the user interface or wireless communication, and collecting user feedback on the acoustic compensation effect; the feedback information includes the user's evaluation of sound clarity, comfort, and volume; D5: Quantify user feedback information and convert it into numerical indicators.

9. A conduction device fitting management system according to claim 8, characterized in that: The calculation gain curve includes: D3.1: Determine the frequency range and number of sampling points of the gain curve; the frequency range covers the audio frequency range that can be processed by the transmission device; D3.2: At each sampling point, calculate the amplitude ratio of the conduction device's output signal to the input signal to obtain the gain value at the sound wave frequency corresponding to the current sampling point; D3.3: Connect the gain values ​​of all sampling points to form a gain curve, and smooth the gain curve.

10. A conducting device, characterized in that: It includes a conducting device body, a data acquisition sensor group, a main control chip, an acoustic compensation module and a user interaction interface; The conducting device body is the main structure of the device and is used to install and fix components; The data acquisition sensor group is arranged on the main body of the conduction device, and is used to collect environmental acoustic characteristic data of the user's environment, the user's physiological index data and the user's motion data, and transmit the collected data to the main control chip; The main control chip receives data transmitted by the data acquisition sensor group and runs the conduction device fitting management system. It dynamically generates compensation parameters based on the real-time fused multi-source fitting data and the scene-physiology-loss three-dimensional adaptive model data, and controls the acoustic compensation module to perform acoustic compensation. The acoustic compensation module receives the compensation parameters transmitted by the main control chip, performs compensation processing on the received audio signal, and outputs the compensated audio signal; The user interaction interface is arranged on the main body of the conduction device, and is used to display the current acoustic compensation effect, receive user feedback information on the acoustic compensation effect, and transmit the feedback information to the main control chip.

Images