A method and system for optimizing a bluetooth hearing aid with adaptive hearing compensation

Abstract

The application discloses a kind of adaptive hearing compensation Bluetooth hearing aid optimization method and system, comprising: S1: collecting user individual hearing curve and the background audio data of left and right ear under each typical scene and respectively pre-processing;S2: extracting the time-frequency fusion feature of each typical scene, generate scene feature vector and construct scene feature vector library;S3: calculate scene perception conversion vector, mutation risk vector, smooth gain vector;S4: calculate binaural gain change vector, binaural synchronization entropy vector, phase alignment factor vector and left and right ear collaborative gain vector;S5: according to left and right ear collaborative gain vector, the background audio data of left and right ear pre-processed under current scene is handled, and output to Bluetooth hearing aid.The application can solve the gain mutation problem generated when existing Bluetooth hearing aid switches scene and the problem of spatial perception distortion caused by binaural collaborative disorder.

Description

Technical Field

[0001] This invention relates to the field of hearing aid technology, and in particular to an adaptive hearing compensation Bluetooth hearing aid headphone optimization method and system. Background Technology

[0002] With the aging population and the continued impact of noise on hearing health, hearing aids are playing an increasingly important role in improving the quality of life for people with hearing impairments. Bluetooth hearing aids, as smart wearable devices integrating modern wireless communication technology and hearing compensation, not only achieve seamless connectivity with smartphones and other terminals, but also use digital signal processing technology to provide personalized gain compensation for users' hearing loss. Traditional hearing aid technologies typically rely on static audiograms for fixed gain settings, or employ simple environmental classification algorithms (such as signal-to-noise ratio detection and spectral feature matching) to coarsely identify noise, speech, and music scenarios, adjusting amplification parameters accordingly. However, these methods often struggle to achieve fine adaptive adjustment in complex and changing real-world auditory environments. Sudden changes in gain parameters, especially during scene transitions, can lead to noticeable "clicking" or sound jumps, causing auditory discomfort and even auditory fatigue. Furthermore, existing systems generally suffer from asynchronicity in gain adjustment between the left and right ears in binaural collaborative processing. Due to the lack of efficient binaural information interaction mechanisms and collaborative control strategies, when one ear responds to environmental changes before the other, it causes binaural gain asymmetry, which in turn interferes with the brain's judgment of sound source spatial localization, leading to spatial perception confusion, weakening the stereo auditory experience, and even affecting user walking safety. These problems severely restrict the natural listening experience and wearing comfort of Bluetooth hearing aids in real-life scenarios, urgently requiring an optimized method and system architecture that can achieve smooth gain transition and binaural synchronous adaptive adjustment. Summary of the Invention

[0003] In view of this, the present invention provides an adaptive hearing compensation Bluetooth hearing aid headphone optimization method and system, which aims to solve the problem of gain abrupt changes when scene switching in existing Bluetooth hearing aid headphones and the problem of spatial perception distortion caused by binaural coordination disorder.

[0004] An optimization method for adaptive hearing compensation Bluetooth hearing aid headphones, comprising:

[0005] S1: Collect the user's personalized hearing curve and the background audio data of the left and right ears in various typical scenarios, and preprocess them respectively to obtain the personalized hearing spectrum vector and the preprocessed background audio data of the left and right ears in various typical scenarios;

[0006] S2: Based on the preprocessed background audio data of the left and right ears and the individual hearing map in each typical scenario, extract the time-domain transformation features and time-frequency fusion features of each typical scenario, generate scene feature vectors and build a scene feature vector library;

[0007] S3: Based on the current scenario, extract the corresponding scenario feature vectors from the scenario feature vector library using the index, and use them as the current scenario feature vector and the target scenario feature vector, and calculate the scenario perception conversion vector; then combine the personalized hearing atlas vector to calculate the mutation risk vector, and then calculate the smoothing gain vector through the risk adaptive smoothing mechanism.

[0008] The calculation process of the mutation risk vector includes: extracting the scene feature vectors of the current and target from the scene feature vector library, and calculating the scene perception conversion vector by combining the difference and cosine similarity; linearly combining the personalized hearing atlas vector with the scene feature vectors of the current and target respectively to calculate the scene gain vectors of the current and target; taking the absolute value of the difference between the scene gain vectors of the current and target and performing a Hadamard product with the hearing sensitivity to obtain the mutation risk vector; wherein the hearing sensitivity is the reciprocal of the personalized hearing atlas vector plus a non-zero constant.

[0009] S4: Based on the preprocessed background audio data of the left and right ears in the current scene and the smoothed gain vector, calculate the binaural gain change vector and binaural synchronization entropy vector; then calculate the phase alignment factor vector and the left and right ear cooperative gain vector.

[0010] S5: Based on the left and right ear collaborative gain vectors, process the pre-processed background audio data of the left and right ears in the current scene, and output it to the Bluetooth hearing aid headphones.

[0011] Furthermore, step S1 also includes:

[0012] S11: Using a hearing test device in a soundproof environment, perform multi-frequency threshold testing on the user, collect the user's personalized hearing curve, and perform bandpass filtering and outlier removal operations to obtain a personalized hearing spectrum vector; the dimension of the personalized hearing spectrum vector is the number of frequency bands, and each element in the vector represents the hearing loss threshold at that frequency;

[0013] S12: Collect background audio data of the left and right ears in various typical scenarios through the dual microphone array of Bluetooth hearing aid headphones. The typical scenarios include quiet indoor environment, street traffic environment, multi-person conversation environment, music playback environment and outdoor wind noise environment.

[0014] S13: Perform adaptive noise reduction and energy normalization operations on the background audio data of the left and right ears in each typical scenario to obtain the preprocessed background audio data of the left and right ears in each typical scenario.

[0015] Furthermore, step S2 also includes:

[0016] S21: Based on the preprocessed background audio data of the left and right ears and the individual hearing map in each typical scenario, extract the temporal transformation features of each typical scenario. The calculation method is as follows:

[0017]

[0018] in, Let i represent the temporal transformation features of the i-th typical scene, where i is the index of the first typical scene. The kernel size is Convolutional layer For Mel filter banks, For modulo operation, It is a short-time Fourier transform. For splicing operations, These are the preprocessed background audio data for the left and right ears in the i-th typical scenario, respectively. For frequency-wise dot product, It is the frequency response function. For individual auditory spectrogram vectors;

[0019] S22: Based on the time-domain transformation characteristics of each typical scenario, extract the time-frequency fusion features of each typical scenario. The calculation method is as follows:

[0020]

[0021] in, Let i be the time-frequency fusion feature of the i-th typical scenario. It is a multilayer perceptron. This is the discrete cosine transform.

[0022] S23: Based on the time-frequency fusion characteristics of each typical scenario, generate a scene feature vector. The calculation method is as follows:

[0023]

[0024]

[0025]

[0026] in, Let be the initial feature vector of the i-th typical scenario. For adaptive pooling layer, Let be the dynamic threshold parameter for the i-th typical scenario. Let be the scene feature vector of the i-th typical scene. For ReLU functions;

[0027] S24: Construct a scene feature vector library V by concatenating the scene feature vectors of each typical scenario.

[0028] Furthermore, step S3 also includes:

[0029] The frequency band difference intensity is calculated by taking the dot product of the frequency band difference intensity weight matrix and the scene perception conversion vector, and then taking the absolute value. The smoothing factor is obtained by exponentially multiplying the dot product of the mutation risk vector and the frequency band difference intensity, adding one, and then taking the reciprocal. The smoothing factor and the scene gain change vector are then multiplied by the Hadamard product and added to the current scene gain vector to obtain the smoothed gain vector.

[0030] Furthermore, step S3 also includes:

[0031] S31: Based on the current scene, extract the corresponding scene feature vectors from the scene feature vector library using the index, and use them as the current scene feature vector and the target scene feature vector. Then, calculate the scene perception transformation vector. The calculation method is as follows:

[0032]

[0033]

[0034]

[0035] in, This is the feature vector of the current scene. Index the scene feature vector library. For the second scene index, For the target scene feature vector, For scene-aware transformation vectors, For cosine similarity calculation, For Hadamah accumulation;

[0036] S32: Based on the current scene feature vector, the target scene feature vector, and the individual hearing atlas vector, calculate the mutation risk vector through hearing sensitivity mapping. The calculation method is as follows:

[0037]

[0038]

[0039]

[0040]

[0041]

[0042] in, This is the current scene gain vector. This is the gain weight matrix for the current scene. This is the current scene gain bias vector. The target scene gain vector. The target scene gain weight matrix, This is the target scene gain bias vector. This is the scene gain change vector. For hearing sensitivity, It is a non-zero constant. , This is the mutation risk vector. To take the absolute value;

[0043] S33: Based on the scene-aware transformation vector and the mutation risk vector, a smoothing gain vector is calculated using a risk-adaptive smoothing mechanism. The calculation method is as follows:

[0044]

[0045]

[0046]

[0047] in, For frequency band difference intensity, This is the frequency band difference intensity weight matrix. As a smoothing factor, For exponential operations of the natural constant, This is the smoothing gain vector.

[0048] It should be further explained that the risk adaptive smoothing mechanism, as a core feature, quantifies the degree of scene change based on the scene perception transformation vector and assesses the level of auditory discomfort risk caused by scene change based on the mutation risk vector, thereby driving the adaptive adjustment of the smoothing factor.

[0049] Specifically, when scenes change rapidly, the mechanism automatically increases the smoothing factor to prolong the gain transition time, effectively suppressing drastic gain fluctuations caused by sudden scene changes and avoiding auditory shock and discomfort. When scene changes are relatively gentle and the change intensity represented by the scene perception conversion vector is weak, the sudden change risk vector will remain at a low level. At this time, the smoothing factor will automatically decrease, shortening the transition response time, so that the gain can quickly converge to the optimal configuration of the target scene, ensuring the timeliness and clarity of auditory perception.

[0050] Compared to traditional algorithms that directly apply gain to the target scene, this invention not only eliminates the instantaneous jump in gain value, significantly reducing the probability of auditory discomfort, but also improves the continuity and naturalness of the audio transition process, avoiding audio distortion and perceptual discontinuity caused by sudden changes. Thus, it achieves precise adaptation to the dynamic characteristics of the scene in complex and ever-changing real-world application environments, ensuring stable maintenance of gain adjustment comfort and robustness in various hearing scenarios, demonstrating stronger environmental adaptability and personalized user response effects than traditional methods.

[0051] Furthermore, in step S4, the calculation process of the binaural gain change vector includes:

[0052] Based on the preprocessed background audio data of the left and right ears in the current scene and the smoothed gain vector, the joint spectral features of the two ears are obtained through short-time Fourier transform and convolution processing. The joint spectral features of the two ears and the smoothed gain vector are mapped to the Hadamard product through the Sigmoid function to obtain the fused latent feature vector. The Hadamard product result of the fused latent feature vector and the smoothed gain vector is linearly transformed to obtain the two-ear gain change vector.

[0053] Furthermore, step S4 also includes:

[0054] S41: Based on the preprocessed background audio data of the left and right ears in the current scene and the smoothed gain vector, calculate the binaural gain change vector and the binaural synchronization entropy vector. The calculation method is as follows:

[0055]

[0056]

[0057]

[0058]

[0059]

[0060] in, The combined spectral characteristics of both ears It is a convolutional layer. These are the preprocessed background audio data for the left and right ears in the current scene, respectively. To fuse latent feature vectors, For Hadama accumulation, For the Sigmoid function, To fuse the latent feature weight matrix, For smoothing the gain vector, This is the binaural gain change vector. This is the weighting matrix for the binaural gain variation. This is the bias vector for the binaural gain variation. This is the frequency band-level synchronization probability vector. This is the binaural synchronization entropy vector. For natural logarithm operations It is a non-zero constant;

[0061] S42: Calculate the phase alignment factor vector based on the binaural synchronization entropy vector. The calculation method is as follows:

[0062]

[0063]

[0064] in, For frequency band stability weights, For exponential operations of the natural constant, As a smoothing factor, This is the phase alignment factor vector;

[0065] S43: Calculate the left and right ear co-gain vectors based on the phase alignment factor vector, binaural gain change vector, and smoothed gain vector. The calculation method is as follows:

[0066]

[0067]

[0068]

[0069]

[0070] in, This is the gain correction vector for the left ear. This is the right ear gain correction vector. The left ear cooperative gain vector, This is the right ear cooperative gain vector.

[0071] It should be further explained that step S4 of the present invention dynamically evaluates the synchronization of the left and right ear audio signals at the frequency band level through the binaural synchronization entropy vector, and generates a phase alignment factor vector accordingly to achieve adaptive allocation of the gain correction direction and intensity of the left and right ears.

[0072] Specifically, when there is phase mismatch in the background audio of the left and right ears in the current scenario, the difference captured by the combined spectral features of the two ears after transformation is converted into a two-ear gain change vector. The two-ear synchronization entropy vector reflects the synchronization stability of each frequency band. When the entropy value is high, it indicates poor synchronization and a high risk of perceptual conflict. At this time, the phase alignment factor vector automatically reduces the response intensity, suppresses excessive gain correction, and avoids the introduction of new auditory interference. For frequency bands with low entropy values ​​and high consistency of two-ear signals, the adjustment effect of the phase alignment factor is enhanced, which promotes the rapid alignment of the left and right ear collaborative gain vectors and improves the spatial auditory fusion effect.

[0073] Compared to the previous method that only adjusted the gain of one ear independently, this invention fully considers the sensitivity of the binaural hearing system to phase consistency and gain balance, and effectively alleviates the spatial perception shift caused by the asymmetry of the gain of the left and right ears.

[0074] Addressing the challenges often encountered in real-world hearing scenarios, such as frequent changes in sound source direction, asymmetrical distribution of background noise, and individual differences in binaural response, this invention achieves consistency and coordination of binaural perception during gain adjustment by real-time modeling of binaural synchronization entropy and frequency band-level control of phase alignment factors. This significantly enhances the naturalness of hearing and spatial localization ability in complex acoustic environments.

[0075] Furthermore, step S5 also includes:

[0076] S51: For the pre-processed background audio data of the left and right ears in the current scene, the signal amplitude is adjusted by the digital signal processor of the Bluetooth hearing aid according to the left and right ear cooperative gain vector, and then the signal is reconstructed by inverse fast Fourier transform to obtain the left and right ear output audio signals.

[0077] S52: Based on the audio signals output from the left and right ears, the audio encoder encapsulates them into a standard audio format to obtain binaural collaborative output audio, which is then output to Bluetooth hearing aid headphones via the Bluetooth audio transmission protocol.

[0078] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0079] (1) To address the problem that existing Bluetooth hearing aids are prone to gain abrupt changes and binaural coordination mismatch leading to spatial perception distortion when switching scenes, this invention first constructs a scene feature vector library containing personalized hearing map vectors and background audio of typical scenes to achieve accurate indexing and difference quantification of the current and target scenes; secondly, it designs a risk adaptive smoothing mechanism jointly driven by scene perception conversion vector and abrupt change risk vector to dynamically adjust the gain transition process, suppressing gain abrupt changes while taking into account response speed; finally, it designs binaural synchronization entropy vector and phase alignment factor vector to perform collaborative correction of the smoothed gain vector under the guidance of binaural gain change vector, realizing phase alignment and spatial consistency optimization of left and right ear gains, significantly improving the auditory comfort, naturalness and spatial perception ability of hearing aids in complex acoustic environments.

[0080] (2) This invention proposes a risk-adaptive smoothing mechanism, which quantifies the degree of scene change by scene perception transformation vector and assesses the risk of auditory discomfort caused by individual differences in hearing sensitivity by combining mutation risk vector, thereby realizing dynamic adjustment of the smoothing factor; when the scene changes rapidly and the intensity of change is large, the smoothing factor is automatically increased to prolong the gain transition time and effectively suppress the auditory impact caused by gain mutation; when the scene change is gentle and the risk is low, the smoothing factor is reduced to speed up the gain response speed and ensure the clarity and timeliness of auditory perception; compared with the traditional direct gain method, this mechanism improves the continuity and naturalness of the transition process while suppressing gain jumps and reducing auditory discomfort, taking into account both comfort and responsiveness, better adapting to complex and ever-changing hearing scenarios, and significantly enhancing the environmental adaptability and user experience of hearing aids in personalized auditory compensation.

[0081] (3) The present invention designs a binaural synchronization entropy vector to dynamically evaluate the synchronization of the left and right ear audio signals at the frequency band level and generate a phase alignment factor vector to achieve adaptive allocation of gain correction. When the binaural signal synchronization is poor and the entropy value is high, the phase alignment factor response intensity is automatically reduced to suppress over-correction and avoid auditory interference. When the synchronization is high and the entropy value is high, the adjustment effect is enhanced to promote the rapid alignment of the left and right ear cooperative gain and improve the spatial auditory fusion effect. Compared with the traditional single-ear independent gain adjustment method, the present invention fully considers the consistency of phase and gain balance of binaural hearing, effectively alleviates the spatial perception shift caused by gain asymmetry, and significantly improves the naturalness of auditory perception and spatial positioning accuracy in complex environments. Attached Figure Description

[0082] Figure 1 A flowchart illustrating an adaptive hearing compensation Bluetooth hearing aid headphone optimization method provided by the present invention;

[0083] Figure 2 A visual comparison of preprocessed audio in two typical scenarios provided by this invention;

[0084] Figure 3 A visual comparison of the time-frequency fusion features in two typical scenarios provided by this invention;

[0085] Figure 4 The structural diagram of the multilayer perceptron in step S22 of this invention is provided. Detailed Implementation

[0086] The present invention will be further described below with reference to the accompanying drawings, but this is not intended to limit the present invention in any way. Any modifications or substitutions made based on the teachings of the present invention shall fall within the protection scope of the present invention.

[0087] Example 1: An optimization method for adaptive hearing compensation Bluetooth hearing aid headphones, such as... Figure 1 As shown, it includes the following steps:

[0088] S1: Collect the user's personalized hearing curve and background audio data for the left and right ears in various typical scenarios, and preprocess them separately to obtain personalized hearing spectrum vectors and preprocessed background audio data for the left and right ears in various typical scenarios, including:

[0089] S11: Using a hearing test device in a soundproof environment, perform multi-frequency threshold testing on the user, collect the user's personalized hearing curve, and perform bandpass filtering and outlier removal operations to obtain a personalized hearing spectrum vector; the dimension of the personalized hearing spectrum vector is the number of frequency bands, and each element in the vector represents the hearing loss threshold at that frequency;

[0090] S12: Collect background audio data of the left and right ears in various typical scenarios through the dual microphone array of Bluetooth hearing aid headphones. The typical scenarios include quiet indoor environment, street traffic environment, multi-person conversation environment, music playback environment and outdoor wind noise environment.

[0091] S13: Perform adaptive noise reduction and energy normalization operations on the background audio data of the left and right ears in various typical scenarios to obtain preprocessed background audio data of the left and right ears in each typical scenario; among them, the background audio data of the left and right ears in multi-person dialogue environment and music playback environment are as follows: Figure 2 As shown;

[0092] S2: Based on the preprocessed background audio data of the left and right ears and individual hearing maps in each typical scenario, extract the temporal transformation features and time-frequency fusion features of each typical scenario, generate scene feature vectors, and construct a scene feature vector library, including:

[0093] S21: Based on the preprocessed background audio data of the left and right ears and the individual hearing map in each typical scenario, extract the temporal transformation features of each typical scenario. The calculation method is as follows:

[0094]

[0095] in, Let i represent the temporal transformation features of the i-th typical scene, where i is the index of the first typical scene. The kernel size is Convolutional layer For Mel filter banks, For modulo operation, It is a short-time Fourier transform. For splicing operations, These are the preprocessed background audio data for the left and right ears in the i-th typical scenario, respectively. For frequency-wise dot product, It is the frequency response function. For individual auditory spectrogram vectors;

[0096] S22: Based on the time-domain transformation characteristics of each typical scenario, extract the time-frequency fusion features of each typical scenario, such as... Figure 3 As shown, the calculation method is as follows:

[0097]

[0098] in, Let i be the time-frequency fusion feature of the i-th typical scenario. This is a multilayer perceptron. For example... Figure 4 As shown, This is the discrete cosine transform.

[0099] S23: Based on the time-frequency fusion characteristics of each typical scenario, generate a scene feature vector. The calculation method is as follows:

[0100]

[0101]

[0102]

[0103] in, Let be the initial feature vector of the i-th typical scenario. For adaptive pooling layer, Let be the dynamic threshold parameter for the i-th typical scenario. Let be the scene feature vector of the i-th typical scene. For ReLU functions;

[0104] S24: Construct a scene feature vector library V by concatenating the scene feature vectors of each typical scenario.

[0105] Specifically, for different scenarios where human speech is the primary mode of communication (such as multi-person dialogue environments with relatively noisy backgrounds), the changes in sound are complex, and traditional methods struggle to simultaneously preserve speech details and accurately represent overall frequency features. Therefore, this invention also provides an adaptive weighted time-frequency fusion feature calculation method to replace step S22. The calculation method is as follows:

[0106]

[0107]

[0108]

[0109]

[0110] in, For the frequency domain transform features of the i-th typical scenario, Let be the temporal weight matrix for the i-th typical scenario. For the Sigmoid function, The kernel size is Convolutional layer Let be the frequency domain weight matrix for the i-th typical scenario. Let i be the time-frequency fusion feature of the i-th typical scenario. For Hadama accumulation, It is the inverse discrete cosine transform.

[0111] S3: Based on the current scenario, extract the corresponding scene feature vectors from the scene feature vector library using the index, and use them as the current scene feature vector and the target scene feature vector, and calculate the scene perception conversion vector; then combine the personalized hearing atlas vector to calculate the mutation risk vector, and then calculate the smoothing gain vector through the risk adaptive smoothing mechanism, including:

[0112] S31: Based on the current scene, extract the corresponding scene feature vectors from the scene feature vector library using the index, and use them as the current scene feature vector and the target scene feature vector. Then, calculate the scene perception transformation vector. The calculation method is as follows:

[0113]

[0114]

[0115]

[0116] in, This is the feature vector of the current scene. Index the scene feature vector library. This is the index for the second typical scenario. For the target scene feature vector, For scene-aware transformation vectors, For cosine similarity calculation, For Hadamah accumulation;

[0117] S32: Based on the current scene feature vector, the target scene feature vector, and the individual hearing atlas vector, calculate the mutation risk vector through hearing sensitivity mapping. The calculation method is as follows:

[0118]

[0119]

[0120]

[0121]

[0122]

[0123] in, This is the current scene gain vector. This is the gain weight matrix for the current scene. This is the current scene gain bias vector. The target scene gain vector. The target scene gain weight matrix, This is the target scene gain bias vector. This is the scene gain change vector. For hearing sensitivity, It is a non-zero constant. , This is the mutation risk vector. To take the absolute value;

[0124] S33: Based on the scene-aware transformation vector and the mutation risk vector, a smoothing gain vector is calculated using a risk-adaptive smoothing mechanism. The calculation method is as follows:

[0125]

[0126]

[0127]

[0128] in, For frequency band difference intensity, This is the frequency band difference intensity weight matrix. As a smoothing factor, For exponential operations of the natural constant, This is the smoothing gain vector.

[0129] For example, when a user switches from a multi-person conversation environment to an outdoor wind noise environment, due to the large difference in scene acoustics and the individual's decreased hearing sensitivity in the mid-to-high frequencies, the product of the mutation risk vector and the intensity of frequency band difference increases significantly. This causes α to decrease to around 0.3 in key frequency bands (such as 2–4kHz), and the gain adjustment process is significantly delayed, effectively suppressing auditory impact. However, when a user switches between similar scenes (such as from a multi-person conversation environment to a street traffic environment), the scene perception change is weak and the difference in hearing sensitivity is small. The product term is relatively small overall, and α is close to or above 0.8. The gain quickly transitions to the target value, ensuring timely response.

[0130] S4: Based on the preprocessed background audio data of the left and right ears in the current scene and the smoothed gain vector, calculate the binaural gain change vector and the binaural synchronization entropy vector; then calculate the phase alignment factor vector and the left and right ear cooperative gain vector, including:

[0131] S41: Based on the preprocessed background audio data of the left and right ears in the current scene and the smoothed gain vector, calculate the binaural gain change vector and the binaural synchronization entropy vector. The calculation method is as follows:

[0132]

[0133]

[0134]

[0135]

[0136]

[0137] in, The combined spectral characteristics of both ears It is a convolutional layer. These are the preprocessed background audio data for the left and right ears in the current scene, respectively. To fuse latent feature vectors, For Hadama accumulation, For the Sigmoid function, To fuse the latent feature weight matrix, For smoothing the gain vector, This is the binaural gain change vector. This is the weighting matrix for the binaural gain variation. This is the bias vector for the binaural gain variation. This is the frequency band-level synchronization probability vector. This is the binaural synchronization entropy vector. For natural logarithm operations It is a non-zero constant;

[0138] S42: Calculate the phase alignment factor vector based on the binaural synchronization entropy vector. The calculation method is as follows:

[0139]

[0140]

[0141] in, For frequency band stability weights, For exponential operations of the natural constant, As a smoothing factor, This is the phase alignment factor vector;

[0142] S43: Calculate the left and right ear co-gain vectors based on the phase alignment factor vector, binaural gain change vector, and smoothed gain vector. The calculation method is as follows:

[0143]

[0144]

[0145]

[0146]

[0147] in, This is the gain correction vector for the left ear. This is the right ear gain correction vector. The left ear cooperative gain vector, This is the right ear cooperative gain vector.

[0148] S5: Based on the left and right ear collaborative gain vectors, process the pre-processed background audio data of the left and right ears in the current scene, and output it to the Bluetooth hearing aid headphones, including:

[0149] S51: For the pre-processed background audio data of the left and right ears in the current scene, the signal amplitude is adjusted by the digital signal processor of the Bluetooth hearing aid according to the left and right ear cooperative gain vector, and then the signal is reconstructed by inverse fast Fourier transform to obtain the left and right ear output audio signals.

[0150] S52: Based on the audio signals output from the left and right ears, the audio encoder encapsulates them into a standard audio format to obtain binaural collaborative output audio, which is then output to Bluetooth hearing aid headphones via the Bluetooth audio transmission protocol.

[0151] Example 2: The present invention also discloses an adaptive hearing compensation Bluetooth hearing aid headphone optimization system, comprising:

[0152] User hearing data and scenario data acquisition module: Collects personalized hearing curves of users and background audio data of the left and right ears in various typical scenarios, and performs preprocessing on each to obtain personalized hearing spectrum vectors and preprocessed background audio data of the left and right ears in various typical scenarios;

[0153] Scene feature vector library construction module: Based on the preprocessed background audio data of the left and right ears and the individual hearing map in each typical scene, extract the time domain transformation features and time-frequency fusion features of each typical scene, generate scene feature vectors and construct the scene feature vector library;

[0154] Smoothing Gain Module: Based on the current scene, the module extracts the corresponding scene feature vectors from the scene feature vector library using an index, which are used as the current scene feature vector and the target scene feature vector, and calculates the scene perception conversion vector; then, it combines the personalized hearing atlas vector to calculate the mutation risk vector, and finally calculates the smoothing gain vector through a risk adaptive smoothing mechanism.

[0155] Left and right ear collaborative gain module: Based on the preprocessed background audio data of the left and right ears in the current scene and the smoothed gain vector, calculate the binaural gain change vector and binaural synchronization entropy vector; then calculate the phase alignment factor vector and the left and right ear collaborative gain vector;

[0156] Audio output module: Based on the left and right ear collaborative gain vectors, it processes the pre-processed background audio data of the left and right ears in the current scene and outputs it to the Bluetooth hearing aid headphones.

[0157] It should be noted that the sequence numbers of the above embodiments of the present invention are merely for descriptive purposes and do not represent the superiority or inferiority of the embodiments. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, apparatus, article, or method. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0158] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0159] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A method for optimizing Bluetooth hearing aid headphones with adaptive hearing compensation, characterized in that, Includes the following steps: S1: Collect the user's personalized hearing curve and the background audio data of the left and right ears in various typical scenarios, and preprocess them respectively to obtain the personalized hearing spectrum vector and the preprocessed background audio data of the left and right ears in various typical scenarios; S2: Based on the preprocessed background audio data of the left and right ears and the individual hearing map in each typical scenario, extract the time-domain transformation features and time-frequency fusion features of each typical scenario, generate scene feature vectors and build a scene feature vector library; S3: Based on the current scenario, extract the corresponding scenario feature vectors from the scenario feature vector library using the index, and use them as the current scenario feature vector and the target scenario feature vector, and calculate the scenario perception conversion vector; then combine the personalized hearing atlas vector to calculate the mutation risk vector, and then calculate the smoothing gain vector through the risk adaptive smoothing mechanism. The calculation process of the mutation risk vector includes: extracting the scene feature vectors of the current and target from the scene feature vector library, and calculating the scene perception conversion vector by combining the difference and cosine similarity; linearly combining the personalized hearing atlas vector with the scene feature vectors of the current and target respectively to calculate the scene gain vectors of the current and target; taking the absolute value of the difference between the scene gain vectors of the current and target and performing a Hadamard product with the hearing sensitivity to obtain the mutation risk vector; wherein the hearing sensitivity is the reciprocal of the personalized hearing atlas vector plus a non-zero constant. S4: Based on the preprocessed background audio data of the left and right ears in the current scene and the smoothed gain vector, calculate the binaural gain change vector and binaural synchronization entropy vector; then calculate the phase alignment factor vector and the left and right ear cooperative gain vector. S5: Based on the left and right ear collaborative gain vectors, process the pre-processed background audio data of the left and right ears in the current scene, and output it to the Bluetooth hearing aid headphones.

2. The adaptive hearing compensation Bluetooth hearing aid headphone optimization method according to claim 1, characterized in that, Step S1 includes: S11: Using a hearing test device in a soundproof environment, perform multi-frequency threshold testing on the user, collect the user's personalized hearing curve, and perform bandpass filtering and outlier removal operations to obtain a personalized hearing spectrum vector; the dimension of the personalized hearing spectrum vector is the number of frequency bands, and each element in the vector represents the hearing loss threshold at that frequency; S12: Collect background audio data of the left and right ears in various typical scenarios through the dual microphone array of Bluetooth hearing aid headphones. The typical scenarios include quiet indoor environment, street traffic environment, multi-person conversation environment, music playback environment and outdoor wind noise environment. S13: Perform adaptive noise reduction and energy normalization operations on the background audio data of the left and right ears in each typical scenario to obtain the preprocessed background audio data of the left and right ears in each typical scenario.

3. The adaptive hearing compensation Bluetooth hearing aid headphone optimization method according to claim 1, characterized in that, Step S2 includes: S21: Based on the preprocessed background audio data of the left and right ears and the individual hearing map in each typical scenario, extract the temporal transformation features of each typical scenario. The calculation method is as follows: in, Let i represent the temporal transformation features of the i-th typical scene, where i is the index of the first typical scene. The kernel size is Convolutional layer For Mel filter banks, For modulo operation, It is a short-time Fourier transform. For splicing operations, These are the preprocessed background audio data for the left and right ears in the i-th typical scenario, respectively. For frequency-wise dot product, It is the frequency response function. For individual auditory spectrogram vectors; S22: Based on the time-domain transformation characteristics of each typical scenario, extract the time-frequency fusion features of each typical scenario. The calculation method is as follows: in, Let i be the time-frequency fusion feature of the i-th typical scenario. It is a multilayer perceptron. This is the discrete cosine transform. S23: Based on the time-frequency fusion characteristics of each typical scenario, generate a scene feature vector. The calculation method is as follows: in, Let be the initial feature vector of the i-th typical scenario. For adaptive pooling layer, Let be the dynamic threshold parameter for the i-th typical scenario. Let be the scene feature vector of the i-th typical scene. For ReLU functions; S24: Construct a scene feature vector library V by concatenating the scene feature vectors of each typical scenario.

4. The adaptive hearing compensation Bluetooth hearing aid headphone optimization method according to claim 3, characterized in that, The risk adaptive smoothing mechanism includes: calculating the frequency band difference intensity by taking the dot product of the frequency band difference intensity weight matrix and the scene perception conversion vector, and then taking the absolute value; exponentially operating the dot product of the mutation risk vector and the frequency band difference intensity, adding one, and then taking the reciprocal to obtain the smoothing factor; performing the Hadamard product of the smoothing factor and the scene gain change vector, and then adding it to the current scene gain vector to obtain the smoothed gain vector. S31: Based on the current scene, extract the corresponding scene feature vectors from the scene feature vector library using the index, and use them as the current scene feature vector and the target scene feature vector. Then, calculate the scene perception transformation vector. The calculation method is as follows: in, This is the feature vector of the current scene. Index the scene feature vector library. This is the index for the second typical scenario. For the target scene feature vector, For scene-aware transformation vectors, For cosine similarity calculation, For Hadamah accumulation; S32: Based on the current scene feature vector, the target scene feature vector, and the individual hearing atlas vector, calculate the mutation risk vector through hearing sensitivity mapping. The calculation method is as follows: in, This is the current scene gain vector. This is the gain weight matrix for the current scene. This is the current scene gain bias vector. The target scene gain vector. The target scene gain weight matrix, This is the target scene gain bias vector. This is the scene gain change vector. For hearing sensitivity, It is a non-zero constant. , This is the mutation risk vector. To take the absolute value; S33: Based on the scene-aware transformation vector and the mutation risk vector, a smoothing gain vector is calculated using a risk-adaptive smoothing mechanism. The calculation method is as follows: in, For frequency band difference intensity, This is the frequency band difference intensity weight matrix. As a smoothing factor, For exponential operations of the natural constant, This is the smoothing gain vector.

5. The adaptive hearing compensation Bluetooth hearing aid headphone optimization method according to claim 3, characterized in that, In step S4, the calculation process of the binaural gain change vector includes: Based on the preprocessed background audio data of the left and right ears in the current scene and the smoothed gain vector, the joint spectral features of the two ears are obtained through short-time Fourier transform and convolution processing. The joint spectral features of the two ears and the smoothed gain vector are mapped to the Hadamard product through the Sigmoid function to obtain the fused latent feature vector. The Hadamard product result of the fused latent feature vector and the smoothed gain vector is linearly transformed to obtain the two-ear gain change vector.

6. The adaptive hearing compensation Bluetooth hearing aid headphone optimization method according to claim 5, characterized in that, Step S4 includes: S41: Based on the preprocessed background audio data of the left and right ears in the current scene and the smoothed gain vector, calculate the binaural gain change vector and the binaural synchronization entropy vector. The calculation method is as follows: in, The combined spectral characteristics of both ears It is a convolutional layer. These are the preprocessed background audio data for the left and right ears in the current scene, respectively. To fuse latent feature vectors, For Hadama accumulation, For the Sigmoid function, To fuse the latent feature weight matrix, For smoothing the gain vector, This is the binaural gain change vector. This is the weighting matrix for the binaural gain variation. This is the bias vector for the binaural gain variation. This is the frequency band-level synchronization probability vector. This is the binaural synchronization entropy vector. For natural logarithm operations It is a non-zero constant; S42: Calculate the phase alignment factor vector based on the binaural synchronization entropy vector. The calculation method is as follows: in, For frequency band stability weights, For exponential operations of the natural constant, As a smoothing factor, This is the phase alignment factor vector; S43: Calculate the left and right ear co-gain vectors based on the phase alignment factor vector, binaural gain change vector, and smoothed gain vector. The calculation method is as follows: in, This is the gain correction vector for the left ear. This is the right ear gain correction vector. The left ear cooperative gain vector, This is the right ear cooperative gain vector.

7. The adaptive hearing compensation Bluetooth hearing aid headphone optimization method according to claim 6, characterized in that, Step S5 includes: S51: For the pre-processed background audio data of the left and right ears in the current scene, the signal amplitude is adjusted by the digital signal processor of the Bluetooth hearing aid according to the left and right ear cooperative gain vector, and then the signal is reconstructed by inverse fast Fourier transform to obtain the left and right ear output audio signals. S52: Based on the audio signals output from the left and right ears, the audio encoder encapsulates them into a standard audio format to obtain binaural collaborative output audio, which is then output to Bluetooth hearing aid headphones via the Bluetooth audio transmission protocol.

8. An adaptive hearing compensation Bluetooth hearing aid headphone optimization system, characterized in that, include: User hearing data and scenario data acquisition module: Collects personalized hearing curves of users and background audio data of the left and right ears in various typical scenarios, and performs preprocessing on each to obtain personalized hearing spectrum vectors and preprocessed background audio data of the left and right ears in various typical scenarios; Scene feature vector library construction module: Based on the preprocessed background audio data of the left and right ears and the individual hearing map in each typical scene, extract the time domain transformation features and time-frequency fusion features of each typical scene, generate scene feature vectors and construct the scene feature vector library; Smoothing Gain Module: Based on the current scene, the module extracts the corresponding scene feature vectors from the scene feature vector library using an index, which are used as the current scene feature vector and the target scene feature vector, and calculates the scene perception conversion vector; then, it combines the personalized hearing atlas vector to calculate the mutation risk vector, and finally calculates the smoothing gain vector through a risk adaptive smoothing mechanism. Left and right ear collaborative gain module: Based on the preprocessed background audio data of the left and right ears in the current scene and the smoothed gain vector, calculate the binaural gain change vector and binaural synchronization entropy vector; then calculate the phase alignment factor vector and the left and right ear collaborative gain vector; Audio output module: Based on the left and right ear collaborative gain vectors, it processes the pre-processed background audio data of the left and right ears in the current scene and outputs it to the Bluetooth hearing aid headphones; To achieve the adaptive hearing compensation Bluetooth hearing aid headphone optimization method as described in any one of claims 1-7.

Images