A method for in-situ hearing threshold assessment and fitting of hearing aids in the external auditory canal

Abstract

This invention discloses a method for in-situ hearing aid threshold assessment and fitting in the external auditory canal, comprising: selecting an assessment environment and standardizing hearing aid wearing; detecting and classifying environmental noise using the built-in microphone of the hearing aid; triggering in-situ adaptive acoustic calibration of the hearing aid, measuring the individual's external auditory canal acoustic transfer function and generating a calibration compensation curve; entering in-situ audiometry mode, cutting off external sound pickup from the hearing aid microphone, and having the hearing aid chip generate a standard audiometry equivalent electrical signal based on the calibration compensation curve and directly inputting it into the signal processing link; conducting pure tone / warp hearing aid threshold testing according to standard frequency points and audiometry specifications, compensating the equivalent electrical signal according to the compensation curve during the test, and automatically performing audiometry quality control; optimizing hearing aid parameters based on the full-band hearing aid threshold test results, and performing in-situ speech recognition threshold verification to complete hearing aid fitting.

Description

Technical Field

[0001] This invention relates to the field of digital signal processing technology for hearing aids, and in particular to a method for in-situ hearing threshold assessment and fitting of hearing aids in the external auditory canal. Background Technology

[0002] In hearing aid fitting, it is usually necessary to assess the hearing threshold after wearing the hearing aid and adjust the hearing aid parameters according to the assessment results so that the user's hearing threshold reaches the expected level.

[0003] Traditional hearing aid threshold assessment and fitting rely on professional soundproof rooms and external standard sound fields (such as sound fields played by external devices like audiometers), making the assessment and fitting process complex. Furthermore, fitting cannot be performed without a soundproof room and professional sound field equipment, limiting the available conditions. Additionally, sound pressure level deviations can occur as the external standard sound field travels through the air to the ear, leading to inaccurate hearing aid threshold assessments and affecting the fitting effect. Summary of the Invention

[0004] This invention provides a method for in-situ hearing threshold assessment and fitting of hearing aids in the external auditory canal, to solve at least one of the above-mentioned problems.

[0005] This invention provides a method for in-situ hearing threshold assessment and fitting of hearing aids in the external auditory canal, including:

[0006] S110. Select an assessment environment and wear hearing aids correctly;

[0007] S120: Detect and classify environmental noise using the built-in microphone of the hearing aid; if the environmental noise test is qualified, proceed directly to S130; if the environmental noise is unqualified but within the range that can be corrected by active noise cancellation, complete active noise cancellation processing and hearing threshold correction preloading through the hearing aid microphone and receiver, then proceed to S130; if the environmental noise exceeds the range that can be corrected by active noise cancellation, end the entire method and prohibit hearing threshold assessment with hearing aid.

[0008] S130, triggers in-situ adaptive acoustic calibration of hearing aids, measures the acoustic transfer function of the individual external auditory canal and generates a calibration compensation curve;

[0009] S140. Enter the in-situ audiometry mode, cut off the external sound pickup of the hearing aid microphone, and have the hearing aid chip generate a standard audiometry equivalent electrical signal according to the calibration compensation curve and directly input it into the signal processing link.

[0010] S150. Conduct pure tone / warp hearing threshold tests according to standard frequency points and audiometry specifications. During the test, the equivalent electrical signal is compensated according to the compensation curve. Audiometry quality control is automatically performed, including invalid feedback screening, silent capture trial false positive verification, and retesting if the false positive rate of a single frequency point exceeds the limit.

[0011] S160: Based on the full-band hearing aid threshold test results, optimize hearing aid parameters and perform in-situ speech recognition threshold verification to complete hearing aid fitting.

[0012] In summary, this invention provides a method for in-situ hearing aid threshold assessment and fitting in the external auditory canal. It establishes a standardized method for hearing aid threshold assessment and fitting that can be used in fitting centers or by users at home. This method includes pre-emptive environmental noise level control, in-situ acoustic calibration, standardized pure-tone / warp audiometry, feedback effectiveness control, false-positive screening, threshold result determination, fitting parameter optimization, and speech recognition threshold verification—a complete set of standardized steps. This method eliminates the need for a professional soundproof room and standard sound field, significantly reducing the requirements for audiologists. Even in a home environment, users can fine-tune the hearing aids themselves without an audiologist, overcoming the environmental and equipment limitations of traditional hearing aid threshold assessment and fitting. Furthermore, the hearing aid chip automatically generates the audiometric signal, avoiding sound pressure level deviations in the external standard sound field during airborne propagation, thus improving the accuracy of hearing aid threshold assessment and fitting.

[0013] Specifically, this embodiment proposes methods for hearing aid microphone noise detection, active noise cancellation correction, and out-of-range testing, filling the gap in existing technologies where audiometry and hearing aid fitting cannot be performed in noisy environments. By adding in-situ calibration of the individual's external auditory canal after hearing aid wearing, acoustic deviations transmitted through the ear canal are compensated, improving the accuracy of hearing aid threshold assessment and hearing aid fitting. At the same time, audiological quality control and speech recognition threshold verification are introduced, forming a complete process of "in-situ calibration → hearing threshold testing → fitting parameter optimization → speech recognition threshold verification," which can be used in professional fitting centers or allows users to independently complete in-situ hearing aid threshold assessment and fitting fine-tuning at home. Attached Figure Description

[0014] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0015] Figure 1 This is a flowchart of a method for in-situ hearing threshold assessment and fitting of hearing aids provided by an embodiment of the present invention;

[0016] Figure 2 This is a schematic diagram of the structure of a sound pressure level deviation prediction model caused by ear canal transmission provided in an embodiment of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0018] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0019] In the description of this invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0020] Figure 1 This is a flowchart illustrating an in-situ hearing aid threshold assessment and fitting method provided by an embodiment of the present invention. This method is completed collaboratively by the fitting personnel / user and the hearing aid threshold assessment and fitting system, and can be applied in scenarios such as fitting centers or users' homes. The hearing aid threshold assessment and fitting system includes the hearing aid and an operating platform outside the hearing aid itself. This platform can be deployed on a desktop computer, an app, etc., for display and interaction with the fitting personnel / user. Figure 1 As shown, the method specifically includes:

[0021] S110. Select an assessment environment and wear hearing aids correctly.

[0022] This step involves pre-assessment preparation and in-situ fitting calibration to ensure that the environment and hearing aid wearing meet the requirements of hearing threshold assessment and hearing aid fitting.

[0023] Optionally, a quiet testing environment with no strong sound reflections should be preferred; the requirement for wearing hearing aids correctly is to fix the position of the earplug / earmold and ensure that the fit is tight and the position is consistent.

[0024] It's worth noting that the assessment environment selection here is only a preliminary one, generally considered to be a quiet environment with no strong sound reflections. More precise noise detection will be conducted using the hearing aids after they are worn.

[0025] S120: Detect and classify environmental noise using the built-in microphone of the hearing aid; if the environmental noise test is qualified, proceed directly to S130; if the environmental noise is unqualified but within the range that can be corrected by active noise cancellation, complete active noise cancellation processing and hearing threshold correction preloading through the hearing aid microphone and receiver, then proceed to S130; if the environmental noise exceeds the range that can be corrected by active noise cancellation, end the entire method and prohibit hearing threshold assessment.

[0026] This embodiment utilizes hearing aids for standardized detection and graded management of environmental noise to accurately determine whether the current environment is suitable for hearing threshold assessment and hearing aid fitting.

[0027] In one specific implementation, after the hearing aid is properly worn, the hearing aid threshold assessment and fitting system controls the hearing aid to activate its built-in microphone, continuously collecting ambient noise data for a sampling duration of ≥3 seconds. The sampling frequency conforms to audiological testing standards, and the equivalent A-weighted sound pressure level and 1 / 3 octave band spectral characteristics are calculated. Then, based on the collected data, the following noise level assessment is performed:

[0028] Level 1 (Environmental Noise Qualified): If the measured A-weighted noise level in the environment is ≤40dB (A), which meets the listening environment standard, the system automatically determines that the noise detection is qualified and directly proceeds to the next step of in-situ adaptive acoustic calibration without the need to activate active noise reduction processing.

[0029] Level 2 (Noise Reduction Correction): If the measured ambient A-weighted noise is >40dB (A), but within the active noise reduction correction range calibrated by the hearing aid manufacturer (default ≤55dB (A)), the system automatically initiates feedforward + feedback composite active noise reduction processing of the hearing aid microphone and receiver to cancel ambient noise in real time. After the noise reduction processing is completed, the system performs a secondary verification of the residual noise level in the ear canal. After confirming that the residual noise meets the audiometry acoustic requirements, it calls the corresponding noise level hearing threshold correction coefficient library calibrated by the manufacturer to complete the preloading of the full-band hearing threshold correction parameters before proceeding to the next step of in-situ acoustic calibration. If the residual noise still cannot meet the audiometry standard after active noise reduction processing, the user is prompted to change the test environment and re-perform the noise test.

[0030] Level 3 (Unable to test): If the measured A-weighted noise exceeds the maximum correctable range of active noise reduction specified by the equipment at the factory (default > 55dB(A)), the system will directly determine that the current environment cannot be used for standardized audiometry, lock the entry point for subsequent audiometry operations, prompt the user to change to a quieter test environment, and restart the testing process. Subsequent operations can only be unlocked after the noise test is qualified or enters the correctable range and noise reduction is completed.

[0031] S130 triggers in-situ adaptive acoustic calibration of the hearing aid, measures the individual's external auditory canal acoustic transfer function, and generates a calibration compensation curve.

[0032] This step involves in-situ acoustic calibration of the hearing aid to ensure that the sound pressure level received at the user's eardrum is consistent with (or has a very small error in) the sound pressure level emitted by the hearing aid, thereby ensuring the accuracy of the hearing threshold assessment.

[0033] In one specific implementation, after triggering the hearing aid's in-situ adaptive acoustic calibration, the hearing aid can play a logarithmic sweep signal to the user. Simultaneously, an in-ear testing instrument (such as a probe microphone) can be used to measure the sound pressure level deviation at the tympanic membrane at each frequency point. This sound pressure level deviation includes deviations caused by ambient noise, wearing position, and sound pressure level transmission through the ear canal. Under conditions where the testing environment meets requirements and the wearing procedure is strictly followed (i.e., fully adhering to the wearing guidelines in S110), the aforementioned sound pressure level deviation primarily originates from sound pressure level deviations caused by sound pressure level transmission through the user's ear canal.

[0034] Then, based on the sound pressure level deviation at the tympanic membrane at each frequency point, an individual ear canal transfer function can be fitted, and a calibration compensation curve can be generated to compensate for the sound pressure level deviation at the tympanic membrane. Optionally, the individual ear canal transfer function can be the frequency domain ratio of the sound pressure at the hearing aid's output point to that at the tympanic membrane. It is a unique natural filtering characteristic of the ear canal for each individual, determined by the length, width, curvature, tympanic membrane stiffness, and soft tissue of the ear canal, and varies from person to person. The compensation calibration curve refers to the frequency domain gain curve used to offset the individual ear canal transfer function. That is, when the hearing aid speaker emits sound at the ear canal opening, the ear canal itself naturally amplifies / attenuates various frequencies (ear canal resonance, cavity loss). The sound received by the tympanic membrane is not exactly the same as the speaker output. Therefore, the compensation calibration curve is: how many dB should be compensated for at each frequency to smooth out the ear canal's natural filtering and make the tympanic membrane sound pressure exactly match the fitting target curve.

[0035] S140. Enter the in-situ audiometry mode, cut off the external sound pickup of the hearing aid microphone, and have the hearing aid chip generate a standard audiometry equivalent electrical signal according to the calibration compensation curve and directly input it into the signal processing link.

[0036] This embodiment does not require external devices to generate hearing test signals. Instead, the hearing aid chip generates the original equivalent electrical signals of each frequency point of the standard hearing test, and superimposes the sound pressure level deviation compensation amount of each frequency point in the calibration compensation curve on the original equivalent electrical signals of each frequency point, so that the sound pressure level deviation at the tympanic membrane after calibration compensation (i.e., the sound pressure level at the tympanic membrane relative to the sound pressure level at the sound source of the hearing aid) is ≤±2dB.

[0037] Optionally, the frequency response and sound pressure level deviation after calibration can be checked. If they pass, the audiometry mode will be entered; if they fail, the user will be prompted to re-wear and recalibrate.

[0038] S150. Conduct pure tone / warp hearing threshold tests according to standard frequency points and audiometry specifications. During the test, the equivalent electrical signal is compensated according to the compensation curve, and audiometry quality control is automatically performed, including invalid feedback screening, silent capture trial false positive verification, and retesting if the false positive rate of a single frequency point exceeds the limit.

[0039] This step involves in-situ hearing aid threshold testing and quality control. In one specific implementation, after entering the in-situ audiometry mode, the system automatically shuts down the hearing aid microphone, and the hearing aid chip automatically generates a standard pure tone / warping signal that is directly input into the signal processing link.

[0040] Optionally, testing can be performed in standard frequency order: 1000Hz first, then traversing the entire frequency band in sequence; alternatively, a Hughson-Westlake stepwise method and Bayesian adaptive audiometry rules can be used. During testing, 5%–10% of silent capture trials can be inserted as false positive quality control; a user feedback response time interval is limited, and feedback exceeding this interval is considered invalid and requires retesting; if the false positive rate at a single frequency point exceeds the threshold, the test is terminated and the user is prompted to refocus on the test.

[0041] Optionally, the user feedback response time can be set to a valid range of 100ms to 5000ms, and feedback outside the range is considered invalid; if the false positive rate of a single frequency point is ≥20%, the current frequency point will be terminated and a retest will be prompted.

[0042] Specific hearing test data, quality control data, hearing test procedures, and quality control procedures can all be integrated into the hearing aid system, allowing users to provide feedback on their hearing status.

[0043] S160: Based on the full-band hearing aid threshold test results, optimize hearing aid parameters and perform in-situ speech recognition threshold verification to complete hearing aid fitting.

[0044] Optionally, the system can integrate hearing aid fitting algorithms such as NAL-NL2 or DSL v5. Based on the measured hearing threshold, it can automatically optimize hearing aid parameters such as gain, compression inflection point, and compression ratio using the NAL-NL2 or DSL v5 hearing aid fitting algorithm; and complete the adjustment by writing new prescription parameters.

[0045] Furthermore, in situ speech recognition threshold (SRT) and speech recognition rate tests can be performed to verify the effectiveness of hearing aids. Finally, an evaluation report is generated as a basis for hearing rehabilitation and fitting.

[0046] In summary, this embodiment provides a method for in-situ hearing aid threshold assessment and fitting in the external auditory canal. It establishes a standardized method for hearing aid threshold assessment and fitting that can be used in fitting centers or by users at home. This includes a complete set of standardized steps such as pre-classification and control of environmental noise, in-situ acoustic calibration, standardized pure-tone / warp audiometry, feedback effectiveness control, false-positive screening, threshold result determination, fitting parameter optimization, and speech recognition threshold verification. This method eliminates the need for a professional soundproof room and standard sound field, significantly reducing the requirements for audiologists. Even in a home environment, it can be performed by the user without an audiologist, overcoming the environmental and equipment limitations of traditional hearing aid threshold assessment and fitting. Simultaneously, the hearing aid chip automatically generates the audiometric signal, avoiding sound pressure level deviations in the external standard sound field during airborne propagation, thus improving the accuracy of hearing aid threshold assessment and fitting.

[0047] Specifically, this embodiment proposes methods for hearing aid microphone noise detection, active noise cancellation correction, and out-of-range testing, filling the gap in existing technologies where audiometry and hearing aid fitting cannot be performed in noisy environments. By adding in-situ calibration of the individual's external auditory canal after hearing aid wearing, acoustic deviations transmitted through the ear canal are compensated, improving the accuracy of hearing aid threshold assessment and hearing aid fitting. At the same time, audiological quality control and speech recognition threshold verification are introduced, forming a complete process of "in-situ calibration → hearing threshold testing → fitting parameter optimization → speech recognition threshold verification". This process can be used in professional fitting centers or allows users to independently complete in-situ hearing aid threshold assessment and fitting fine-tuning at home.

[0048] Furthermore, while in-ear testing devices are commonly found in audiology centers, they may not be available in home environments. To address the lack of in-ear testing devices, this embodiment can utilize data from in-ear testing devices to train a neural network, which can then automatically predict the sound pressure level deviation at the user's eardrum instead of using an in-ear testing device.

[0049] In one specific implementation, as described above, under conditions where the testing environment meets the requirements and the hearing aid is worn according to strict specifications, the sound pressure level deviation at the tympanic membrane at each frequency point measured using the in-ear testing instrument is approximately equal to the sound pressure level deviation caused by the transmission through the ear canal at each frequency point. Therefore, the hearing aid threshold assessment and fitting system may also include a cloud server, which can collect in-ear testing instrument data from multiple users under strict hearing aid wearing specifications (assuming the testing environment meets noise requirements) as the sound pressure level deviation caused by the transmission through the ear canal at each frequency point for each user. The sound pressure levels and deviations caused by the transmission through the ear canal at each frequency point for the same user, along with the user's three-dimensional ear canal model, are used as a sample to construct a sample set. This sample set is used to train a neural network. The trained neural network takes a user's three-dimensional ear canal model as input and the sound pressure level deviation caused by the transmission through the ear canal at each frequency point for that user as output. The trained neural network can be applied to assessment and fitting scenarios lacking in-ear testing instruments, automatically predicting the sound pressure level deviation caused by the transmission through the ear canal at each frequency point based on the user's three-dimensional ear canal model.

[0050] Whether hearing aids are worn correctly can be checked by an audiologist at the fitting center. At home or in other settings, users can check for themselves whether they have followed the above guidelines for hearing aid wear; alternatively, users can turn off the hearing aid microphone and check if they can still hear external sounds. If it is sufficiently quiet, it can be considered that the hearing aids are worn correctly.

[0051] Optional, such as Figure 2 As shown, the trained neural network includes a first neural network and multiple second neural networks. When assessing and fitting hearing thresholds in the external auditory canal for new users, after completing S110-S120, S130 includes:

[0052] Step 1: Obtain a three-dimensional ear canal model of the new user under strict hearing aid wearing guidelines.

[0053] The "new user" here refers to a new user encountered after the aforementioned neural network model has been trained. Before fitting a customized hearing aid, a three-dimensional ear canal model of the user has been obtained using three-dimensional scanning technology. The scanned three-dimensional point cloud data, or voxel data, or surface mesh data has been uploaded to the cloud, serving as the data source for predicting the sound pressure level deviation caused by the transmission through the user's ear canal in this embodiment.

[0054] Step 2: Input the three-dimensional ear canal model into the first neural network for classification to obtain the deviation type.

[0055] Here, the deviation type refers to the category of sound pressure level deviation caused by the transmission through the user's ear canal. Due to differences in ear canal length, ear canal structure, etc., the signal loss caused by the transmission of sound signals in the ear canals of different users varies. To reduce the difficulty of processing user differences, this embodiment classifies the diverse sound pressure level deviations collected in actual applications. This step first uses a first neural network to determine the deviation type to which the sound pressure level deviation of the current user belongs.

[0056] Combination Figure 2 After inputting the current user's three-dimensional ear canal model into the first neural network, the first neural network can output the probability of the current user belonging to each deviation type, and the deviation type with the highest probability is the user's deviation type. Figure 2 The table shows N types of deviations.

[0057] Step 3: Invoke the second neural network corresponding to the deviation type, and the reference value of the sound pressure level deviation caused by the ear canal transmission at each frequency point corresponding to the deviation type.

[0058] Combination Figure 2 Each deviation type corresponds to a second neural network, and each deviation type corresponds to a reference value for sound pressure level deviation. Each second neural network takes a three-dimensional ear canal model of the user corresponding to the deviation type as input and outputs the change in the user's sound pressure level deviation relative to the reference value of the corresponding deviation type.

[0059] For example, the reference value for a certain type of deviation includes the sound pressure level deviation of the sound signal at each frequency and sound pressure level:

[0060] (1)

[0061] The output of the second neural network for this type of bias includes:

[0062] (2)

[0063] Where f represents the number of frequency points and n represents the number of sound pressure levels. This represents the sound pressure level deviation at the i-th sound pressure level at the j-th frequency. This indicates the sound pressure level deviation at the j-th frequency and the ith sound pressure level of the input user. The difference is i=1,2,…n,j=1,2,…,f.

[0064] After determining the type of deviation for the current user, this step calls the second neural network corresponding to that type of deviation for further processing.

[0065] Step 4: Input the three-dimensional ear canal model into the second neural network to obtain the difference between the sound pressure level deviation caused by the ear canal transmission at each frequency point of the new user and the reference value; and calculate the sound pressure level deviation caused by the ear canal transmission at each frequency point of the new user based on the difference and the reference value, as the sound pressure level deviation at the tympanic membrane at each frequency point of the new user.

[0066] Adding the difference to the baseline value yields the current user's sound pressure level and deviation. Using the matrix described above as an example, assuming the current user's baseline value and the output of the second neural network respectively include... and Then the sound pressure level deviation of this user is:

[0067] (3)

[0068] in, This represents the sound pressure level deviation at the i-th sound pressure level at the j-th frequency for this user.

[0069] Step 5: Based on the sound pressure level deviation at the tympanic membrane at each frequency point of the new user, fit the individual ear canal transfer function and generate a calibration compensation curve to compensate for the sound pressure level deviation at the tympanic membrane.

[0070] In this embodiment, to reduce the difficulty of predicting sound pressure level deviations caused by ear canal transmission using a neural network model, the sound pressure level deviations are first classified. Users in the same category have the same baseline value for sound pressure level deviation, and their actual sound pressure level deviations are close to the baseline value. Fine-tuning can then be performed using a second neural network based on the baseline value. This model architecture of classification followed by fine-tuning reduces the difficulty of model training and prediction, and improves prediction accuracy.

[0071] Furthermore, in another specific embodiment, the following method can be used to... Figure 2 The neural network model shown is trained as follows:

[0072] Step 1: Cluster the sound pressure level deviations in the training samples, grouping users belonging to the same cluster into the same deviation type, and using the cluster centers as the baseline value for the sound pressure level deviation corresponding to each deviation type. This step clusters the sound pressure level deviation data in each sample based on data characteristics, classifying diverse sample data into several categories. It transforms the learning of data patterns from a large number of samples into learning the data patterns between a few categories, reducing training difficulty and improving training accuracy.

[0073] Meanwhile, although the sound pressure level deviation in the sample is related to the user's ear canal morphology, structure, and size, this embodiment does not cluster the user's three-dimensional ear canal model, but directly clusters the prediction target (sound pressure level deviation). This can directly capture the category features of the prediction target, filter out information in the three-dimensional ear canal that is irrelevant to the sound pressure level deviation, and give full play to the guiding role of the classification results in subsequent predictions.

[0074] Step 2: Train the first neural network using the three-dimensional ear canal model of each user, so that the first neural network can output the deviation type of each user.

[0075] Optionally, the first neural network is trained using the three-dimensional ear canal model of each user, so that the first neural network can output the probability of each user belonging to each deviation type, and the deviation type with the highest probability is the user's deviation type.

[0076] Optionally, during training, the bias type obtained from clustering can be used as the true value. The cross-entropy loss function can be used to constrain the bias type of the model output to be consistent with the true value, thereby completing the training of the first neural network.

[0077] Step 3: For each deviation type: Use the three-dimensional ear canal model of each user in the current deviation type to train the second neural network corresponding to the current deviation type, so that the second neural network can output the difference between the sound pressure level deviation in each sample and the benchmark value corresponding to the current deviation type.

[0078] This step trains a second neural network for each type of deviation using samples from that type. In other words, it does not require learning a universal rule applicable to all users, but only the characteristics of ear canal transmission deviation for each type of user. This further reduces the difficulty of model training and makes it more effective in obtaining sound pressure level deviation data applicable to the current type.

[0079] In one specific implementation, the first neural network and each of the second neural networks may adopt the same backbone network, such as the backbone network structure of point transformer or point CNN, with different prediction layers connected after the backbone network.

[0080] Optionally, each prediction layer can adopt a multilayer perceptron structure. The difference between the prediction layers of the two neural networks is that the prediction layer of the first neural network is a classification network, which outputs the probability of each category and uses a classification loss function (such as cross-entropy); while the prediction layer of the second neural network is a regression prediction network, which outputs the data included in the second sound pressure level deviation and uses a regression loss function (such as L1 loss function or L2 loss function), so that the sound pressure level deviation obtained by back-inferring from the model output tends to be consistent with the actual sound pressure level deviation of the sample.

[0081] Furthermore, the backbone network of each second neural network can use the backbone network that has already been trained by the first neural network. This trained network is already able to learn some features in the three-dimensional ear canal model that are closely related to the sound pressure level deviation. Based on these features, parameter fine-tuning and prediction head adjustment are performed for each type of specific feature, which can more quickly learn the differentiated features of each user and improve the convergence speed and prediction accuracy of model training.

[0082] Specifically, since errors are inevitable in neural network training and prediction, the following situation should be considered during training: If a user obtains multiple bias types with the highest probability in step two (usually two; if more, the probability of each type is less than 0.33, which is unreliable), then the two most probable bias types can be referred to as the first bias type and the second bias type. If the probability difference between these two bias types is less than a first threshold (e.g., 0.01), and the probabilities of both bias types are greater than a second threshold (e.g., 0.45), then the sample is likely to belong to both the first and second bias types. In this case, the sample's affiliation to two clusters in step one is also similar. Therefore, these samples are indeed more prone to misclassification during neural network training and prediction, or in other words, being classified into any category would be considered correct.

[0083] In order to enable the model to have a certain classification tolerance for these samples, that is, to still obtain the correct sound pressure level deviation in the case of classification error or classification boundary simulation, in the training of step three, this embodiment simultaneously inputs the three-dimensional ear canal model of these samples into the second neural network corresponding to the first deviation type and the second neural network corresponding to the second deviation type, and performs joint training on the two second neural networks.

[0084] Specifically, after inputting a user's three-dimensional ear canal model into the second neural network corresponding to the first deviation type, the difference between the sound pressure level deviation caused by the user's ear canal transmission and the benchmark value corresponding to the first deviation type is obtained; after inputting the user's three-dimensional ear canal model into the second neural network corresponding to the second deviation type, the difference between the sound pressure level deviation caused by the user's ear canal transmission and the benchmark value corresponding to the second deviation type is obtained. For ease of distinction and description, this embodiment refers to the benchmark value corresponding to the first deviation type as the first benchmark value, and the difference relative to the first benchmark value as the first difference; the benchmark value corresponding to the second deviation type is referred to as the second benchmark value, and the difference relative to the second benchmark value as the second difference.

[0085] Then, based on the first reference value and the first difference, a value of the sound pressure level deviation caused by the transmission through the user's ear canal is predicted (referred to as the first value); simultaneously, based on the second reference value and the second difference, another value of the sound pressure level deviation caused by the transmission through the user's ear canal is predicted (referred to as the second value). Of course, when the sound pressure level deviation includes multiple data points, both the first and second differences include the difference corresponding to each data point, and the first and second values ​​also include the value of each data point. In this step, two sets of sound pressure level deviation data can be obtained based on the two sets of difference data and the reference values ​​corresponding to the two sets of difference data.

[0086] Finally, by constraining the first and second values ​​to tend to be consistent, the second neural network corresponding to the first deviation type and the second neural network corresponding to the second deviation type are jointly trained. For example, taking the data shown in formulas (1)-(3) as an example, the following loss term can be added to the loss function of each second neural network model. :

[0087] (4)

[0088] in, Indicating the first value , Indicates the second value By constraining the prediction result regions of the two second neural networks to be consistent, easily confused samples that are on the edge of two classes can obtain correct prediction results regardless of which class they are classified into. This gives the first neural network a certain degree of fault tolerance and improves the overall prediction accuracy and stability of the neural network composed of the first neural network and each of the second neural networks.

[0089] In summary, this embodiment, after accumulating a certain number of sound pressure level deviation samples measured by an in-ear testing instrument, uses a two-stage neural network structure—classification followed by adjustment—to directly utilize the three-dimensional ear canal model data inevitably collected from the user during the customization of the hearing aid. This deviation, under the condition that the testing environment meets the requirements and the hearing aid is worn according to strict specifications, can replace the test data from the in-ear testing instrument, thus enabling in-situ adaptive acoustic calibration of the hearing aid even in scenarios lacking an in-ear testing instrument. Specifically, to improve prediction accuracy, this embodiment achieves deviation type labeling through data clustering. By classifying, the task of learning patterns applicable to all samples is transformed into a task of learning differences in deviation types, and a task of learning local patterns applicable to each type of sample. The transformed tasks are easier to implement and more likely to achieve the desired prediction accuracy. Meanwhile, to address potential edge samples in the classification process, this embodiment uses joint training of two types of second neural networks to compensate for possible errors in the classification. This ensures that even with slight errors, the correct sound pressure level deviation can be obtained regardless of the approximate type identified, further improving the accuracy and stability of the user's ear canal transmission distortion prediction, thereby guaranteeing the accuracy of hearing aid threshold assessment and hearing aid fitting.

[0090] It should be noted that all user data involved in this application is information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation portals are provided for users to choose to authorize or refuse.

[0091] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the technical solutions of the embodiments of the present invention.

Claims

1. A method for in-situ hearing threshold assessment and fitting of hearing aids in the external auditory canal, characterized in that, include: S110. Select an assessment environment and wear hearing aids correctly; S120: Environmental noise detection and classification are performed using the built-in microphone of the hearing aid. If the environmental noise test is qualified, proceed directly to S130; if the environmental noise is unqualified but within the range that can be corrected by active noise cancellation, after active noise cancellation processing and hearing threshold correction preloading are completed through the hearing aid microphone and receiver, proceed to S130; if the environmental noise exceeds the range that can be corrected by active noise cancellation, end the entire method and prohibit hearing aid hearing threshold assessment. S130, triggers in-situ adaptive acoustic calibration of hearing aids, measures the acoustic transfer function of the individual external auditory canal and generates a calibration compensation curve; S140. Enter the in-situ audiometry mode, cut off the external sound pickup of the hearing aid microphone, and have the hearing aid chip generate a standard audiometry equivalent electrical signal according to the calibration compensation curve and directly input it into the signal processing link. S150. Conduct pure tone / warp hearing threshold tests according to standard frequency points and audiometry specifications. During the test, the equivalent electrical signal is compensated according to the compensation curve. Audiometry quality control is automatically performed, including invalid feedback screening, silent capture trial false positive verification, and retesting if the false positive rate of a single frequency point exceeds the limit. S160: Based on the full-band hearing aid threshold test results, optimize hearing aid parameters and perform in-situ speech recognition threshold verification to complete hearing aid fitting.

2. The method according to claim 1, characterized in that, In S120: The acceptable threshold for environmental noise is: A-weighted environmental noise ≤ 40 dB; The correctable range for environmental noise is: A-weighted environmental noise in the range of (40 dBA, 55 dB). The prohibited threshold for environmental noise is: A-weighted environmental noise > 55 dB.

3. The method according to claim 1, characterized in that, S130 includes: The hearing aid plays a logarithmic sweep signal; The sound pressure level deviation at the tympanic membrane at various frequencies was measured using an in-ear tester. The sound pressure level deviation included the sound pressure level deviation caused by environmental noise, wearing position, and sound pressure level transmission through the ear canal. Based on the sound pressure level deviation at the tympanic membrane at each frequency point, the individual ear canal transfer function is fitted, and a calibration compensation curve is generated to compensate for the sound pressure level deviation at the tympanic membrane.

4. The method according to claim 3, characterized in that, The direct input signal processing link, in which the hearing aid chip generates a standard audiometric equivalent electrical signal based on the calibration compensation curve and directly inputs it, includes: The hearing aid chip generates the original equivalent electrical signals at each frequency point of the standard hearing test, and the sound pressure level deviation compensation amount of each frequency point in the calibration compensation curve is superimposed on the original equivalent electrical signals at each frequency point.

5. The method according to claim 4, characterized in that, The sound pressure level deviation at the tympanic membrane after compensation is ≤ ±2dB.

6. The method according to claim 1, characterized in that, The process of invalid feedback screening, silent detection of false positives, and retesting when the false positive rate exceeds the limit at a single frequency point includes: The user feedback response time is set to be within the valid range of 100ms to 5000ms; feedback outside this range is considered invalid. Randomly insert 5% to 10% of silent capture trials; If the false positive rate of a single frequency point is ≥20%, the current frequency point will be terminated and a retest will be prompted.

7. The method according to claim 1, characterized in that, S160 includes: The NAL-NL2 or DSL v5 hearing aid fitting algorithm is used to optimize the hearing aid's gain, compression inflection point, and compression ratio. In situ speech recognition threshold (SRT) and speech recognition rate were tested to verify the effectiveness of the hearing aid.

8. The method according to claim 3, characterized in that, After measuring the sound pressure level deviation at the tympanic membrane at each frequency point using an in-ear tester, the method further includes: Under strict and standardized hearing aid wearing conditions, the sound pressure level deviation at the tympanic membrane at each frequency point is taken as the sound pressure level deviation caused by the transmission through the ear canal at each frequency point. The sound pressure level deviations caused by the transmission of sound pressure levels through the ear canal at various frequencies from multiple users are collected and used together with the three-dimensional ear canal models of each user as samples to train a neural network. The trained neural network takes the three-dimensional ear canal model of a user as input and the sound pressure level deviations caused by the transmission of sound pressure levels through the ear canal at various frequencies from that user as output.

9. The method according to claim 8, characterized in that, The trained neural network includes a first neural network and multiple second neural networks; Accordingly, when assessing and fitting hearing thresholds in the external auditory canal for new users, S130 includes: Under strict guidelines for hearing aid wearing, obtain a three-dimensional ear canal model for new users; The three-dimensional ear canal model is input into the first neural network for classification to obtain the deviation type; The second neural network corresponding to the deviation type is invoked, along with the reference value of the sound pressure level deviation caused by the lower ear canal transmission at each frequency point corresponding to the deviation type. The three-dimensional ear canal model is input into the second neural network to obtain the difference between the sound pressure level deviation caused by the ear canal transmission at each frequency point of the new user and the benchmark value. Based on the differences and the benchmark values, the sound pressure level deviation caused by the ear canal transmission at each frequency point of the new user is calculated, and this deviation is taken as the sound pressure level deviation at the tympanic membrane at each frequency point of the new user. Based on the sound pressure level deviation at the tympanic membrane at each frequency point of the new user, the individual ear canal transfer function is fitted, and a calibration compensation curve is generated to compensate for the sound pressure level deviation at the tympanic membrane.

10. The method according to claim 9, characterized in that, With strict adherence to hearing aid wearing guidelines, the in-situ hearing threshold assessment and fitting method for hearing aids in the external auditory canal of the new user can be applied in the home environment.

Images