A hearing aid device based on fusion of air conduction signals and structure conduction signals, a hearing aid device self-sound suppression method, system, terminal and medium

Abstract

The application discloses a hearing aid device based on air conduction signal and structure conduction signal fusion, a hearing aid device self-sound suppression method and system, a terminal and a medium. The hearing aid device comprises a shell assembly, at least one air conduction microphone, at least one structure conduction signal sensing unit, a signal processing module and a receiver. The air conduction microphone obtains an air conduction sound signal containing wearer self-sound and non-wearer speech. The structure conduction signal sensing unit is mechanically coupled with the shell structure, and converts the structure vibration caused by the wearer's sound into a structure conduction signal. The signal processing module constructs fusion input information according to the corresponding relationship of the two signals, distinguishes the two types of speech through a speech processing model, reduces the output proportion of the wearer's self-sound, and at the same time maintains or enhances the non-wearer speech. The application effectively improves the self-sound recognition and selective suppression capability, and reduces the false suppression of external communication speech.

Description

Technical Field

[0001] This invention relates to the technical field of hearing aids and audio signal processing, and in particular to a hearing aid based on the fusion of air conduction signals and structural conduction signals, a hearing aid self-sound suppression method, system, terminal and medium. Background Technology

[0002] Hearing aids typically use one or more air-conducting microphones to collect ambient sound, which is then processed through noise reduction, equalization, and hearing compensation before being output to the wearer via a receiver. When the wearer speaks, their sound travels not only through the air to the microphone but also through the skull, soft tissue, auricle, ear canal, shell contact surface, and structural components of the hearing aid, forming a structural conduction path.

[0003] For closed or semi-closed hearing aid structures, the wearer's voice is prone to a blockage effect, causing the self-sound to be amplified within the ear canal. For open hearing aid structures, the wearer's self-sound also enters the hearing aid processing chain via air conduction, bone conduction, tissue conduction, and device structure coupling. Existing hearing aid algorithms typically rely on one or more air conduction microphones to perform noise reduction, beamforming, automatic gain control, direction determination, or fixed gain adjustment. However, microphones simultaneously receive the wearer's self-sound, non-wearer's speech, and ambient sound, making it difficult to reliably determine whether the current speech is produced by the wearer. Therefore, existing processing methods are prone to problems such as self-sound residue, insufficient suppression, or weakening of external communication speech while suppressing self-sound.

[0004] Some solutions use motion sensors to help determine whether the wearer is speaking, or switch the hearing aid gain after detecting self-sound. However, they usually only use structural conduction signals as trigger information or state control information, and do not make full use of the continuous correspondence between structural conduction signals and air conduction sound signals to jointly estimate the wearer's speech and the non-wearer's speech.

[0005] Therefore, existing technologies still have shortcomings. Summary of the Invention

[0006] To address the aforementioned deficiencies in existing technologies, this invention provides a hearing aid device based on the fusion of air conduction signals and structural conduction signals, a self-sound suppression method for the hearing aid device, a system, a terminal, and a medium. The technical solution adopted by this invention is as follows: In a first aspect, the present invention provides a hearing aid device based on the fusion of air conduction signals and structural conduction signals, the hearing aid device comprising: Housing assembly; At least one air-conducting microphone is used to acquire air-conducted sound signals containing the wearer's own voice and non-wearer's voice; At least one structural conduction signal sensing unit is disposed within the housing assembly and mechanically coupled to a structural component that receives structural vibrations caused by the wearer's voice, for converting structural vibrations and / or structural conduction sound into structural conduction signals; The signal processing module is connected to both the air conduction microphone and the structure conduction signal sensing unit. The receiver is connected to the signal processing module; The signal processing module is configured as follows: Acquire the air-conducted sound signal and the structure-conducted sound signal; The fused input information is constructed based on the correspondence between the air-conducted sound signal and the structure-conducted sound signal; The fused input information is input into the speech processing model to obtain an output result used to distinguish between the wearer's speech and the non-wearer's speech; Based on the output results, the proportion of the wearer's voice in the hearing aid output is reduced, while the non-wearer's voice is maintained or enhanced, and then output through the receiver.

[0007] In one implementation, the structural conductive signal sensing unit includes any one or more of the following: a vibration pickup unit, a piezoelectric vibration sensor, a MEMS accelerometer, a contact microphone, a bone conduction pickup sensor, an inertial sensor, a strain sensor, a force sensor, a vibration velocity sensor, or a displacement sensor.

[0008] In one implementation, the structural conductive signal sensing unit is connected to the structural component by any one of the following methods: bonding, snap-fitting, fastening, pre-compression, elastic pad coupling, or integral molding, so that the structural vibration generated by the wearer's voice is transmitted to the structural conductive signal sensing unit, and the influence of external airborne sound on the output of the structural conductive signal is reduced.

[0009] In one implementation, the structural conductive signal sensing unit is disposed in at least one of the following preset installation areas: the middle part of the upper housing, the front part of the upper housing, the rear part of the upper housing, between the internal support and the upper housing, between the internal support and the lower housing, the center of the internal support, the middle part of the lower housing, the front part of the lower housing, the rear part of the lower housing, the rear part of the hearing aid, the pickup end, the end of the receiver, the inner wall of the upper housing, the inner wall of the lower housing, or an area on the internal support where a mechanical coupling path can be formed; The installation posture, sensitive direction, and base orientation of the structural conductive signal sensing unit are set according to the housing space and the direction of structural vibration transmission.

[0010] In one implementation, the fused input information includes at least one of the following: time-domain waveforms, time-frequency representations, amplitudes, phases, band energy, envelopes, fundamental frequencies, harmonics, channel amplitude differences, channel amplitude ratios, cross-correlation, coherence, time delays, speech activity states, and self-speech confidence levels of the air-conducted acoustic signal and the structure-conducted acoustic signal.

[0011] In one implementation, the speech processing model includes any one or more of the following: rule-based decision model, statistical classification model, adaptive filtering model, source separation model, and neural network model; the speech processing model is used to determine the wearer's self-voice interval or self-voice confidence based on the fusion input information of the air-conducted signal and the structure-conducted signal, and to estimate the wearer's speech component and non-wearer's speech component in the air-conducted sound signal.

[0012] In one implementation, the neural network model includes an input layer, a temporal modeling layer, and an output layer; the input layer is used to receive an input representation obtained from at least one airborne acoustic signal and at least one structural acoustic signal; the temporal modeling layer includes a Long Short-Term Memory (LSTM) network, or any one or more of a recurrent neural network, a gated recurrent unit, a temporal convolutional network, a convolutional neural network, a Transformer network, and a state-space network.

[0013] In one implementation, the output layer employs at least one of the following output methods: Output two signals: wearer speech estimation and non-wearer speech estimation. Output one channel of speech estimation and wearer's speech activity detection results or self-speech confidence; Output the wearer's speech estimate, and obtain the non-wearer's speech estimate from the airborne sound signal based on the wearer's speech estimate; The output corresponds to the time-frequency mask, gain, category label, or category probability of the wearer's speech and the non-wearer's speech, in order to distinguish and control the wearer's speech and the non-wearer's speech separately.

[0014] In one implementation, the signal processing module generates a hearing aid output according to y = α·y_non_wearer + β·y_wearer, where y_non_wearer is the speech estimation of the non-wearer, y_wearer is the speech estimation of the wearer, α is the speech gain coefficient of the non-wearer, and β is the self-voice preservation coefficient of the wearer; α and β are preset fixed values ​​or dynamically adjusted values, respectively, and the dynamically adjusted value is determined based on at least one of self-voice confidence, user self-voice suppression level, input sound level, hearing compensation parameters, wearing status, and usage scenario.

[0015] Secondly, embodiments of the present invention also provide a self-sound suppression method for a hearing aid device based on the fusion of air conduction signals and structural conduction signals, wherein the self-sound suppression method is implemented using any of the above-described solutions using a hearing aid device based on the fusion of air conduction signals and structural conduction signals, and the method includes: Simultaneously acquire at least one airborne acoustic signal and at least one structural acoustic signal; The fused input information is constructed based on the correspondence between the air-conducted sound signal and the structure-conducted sound signal; The fused input information is input into the speech processing model to obtain an output result used to distinguish between the wearer's speech and the non-wearer's speech; Based on the output results, the proportion of the wearer's voice in the hearing aid output is reduced, while the non-wearer's voice is maintained or enhanced, generating a self-sound-suppressed hearing aid signal, which is then output through the receiver.

[0016] In one implementation, the speech processing model obtains wearer speech estimation and non-wearer speech estimation through at least one of the following methods: The output time-frequency masks corresponding to the wearer's voice and the non-wearer's voice are respectively applied to the reference signal; Direct mapping yields two time-frequency representations; Directly generate two time-domain signals; The wearer's own voice component in the airborne sound signal is estimated based on the structure-conducted signal, and all or part of the wearer's own voice component is removed from the airborne sound signal.

[0017] In one implementation, reducing the proportion of the wearer's speech in the hearing aid output based on the output result, while maintaining or enhancing the non-wearer's speech, includes: When the correspondence between the structurally conducted signal and the air-conducted sound signal satisfies the wearer's self-voice determination condition, the self-voice confidence is increased, and the proportion of the wearer's speech component in the hearing aid output is reduced. When the airborne sound signal contains speech but the structure-conducted sound signal does not generate a corresponding response, it is determined to be the speech of a non-wearer, and the output of the non-wearer speech component is maintained or enhanced.

[0018] In one implementation, the speech processing model is trained using training samples that include wearer speech, non-wearer speech, environmental noise, and corresponding structural conduction signals. The training objective is used to constrain at least one of the following: source classification of wearer speech and non-wearer speech, speech activity detection, channel attribution, separation, speech intelligibility, and waveform fidelity.

[0019] Thirdly, embodiments of the present invention also provide a self-sound suppression system for a hearing aid device based on the fusion of air conduction signals and structural conduction signals, wherein the system is used to implement the steps of the self-sound suppression method for a hearing aid device based on the fusion of air conduction signals and structural conduction signals as described in any of the above solutions, the method comprising: A conductive signal acquisition module is used to simultaneously acquire at least one airborne acoustic signal and at least one structural conductive signal; A fusion input construction module is used to construct fusion input information based on the correspondence between the air-conducted sound signal and the structure-conducted sound signal; The speech separation processing module is used to input the fused input information into the speech processing model to obtain an output result that distinguishes the wearer's speech from the non-wearer's speech; The output control module is used to reduce the proportion of the wearer's voice in the hearing aid output according to the output result, while maintaining or enhancing the non-wearer's voice, generating a self-sound-suppressed hearing aid signal, and outputting it through the receiver.

[0020] Fourthly, embodiments of the present invention also provide a terminal, wherein the terminal includes a memory, a processor, and a self-sound suppression program for a hearing aid device based on the fusion of air conduction signals and structural conduction signals, stored in the memory and executable on the processor. When the processor executes the self-sound suppression program for a hearing aid device based on the fusion of air conduction signals and structural conduction signals, it implements the steps of the self-sound suppression method for a hearing aid device based on the fusion of air conduction signals and structural conduction signals in any of the above-described solutions.

[0021] Fifthly, embodiments of the present invention also provide a computer-readable storage medium, wherein the computer-readable storage medium stores a self-sound suppression program for a hearing aid based on the fusion of air conduction signals and structural conduction signals, the self-sound suppression program for a hearing aid based on the fusion of air conduction signals and structural conduction signals implementing the steps of the self-sound suppression method for a hearing aid based on the fusion of air conduction signals and structural conduction signals as described in any of the above-described schemes on the computer-readable storage medium.

[0022] Beneficial Effects: Compared with the prior art, the present invention provides a hearing aid device based on the fusion of air conduction signals and structural conduction signals. The hearing aid device includes: a housing assembly; at least one air conduction microphone for acquiring air conduction sound signals containing the wearer's own voice and non-wearer's voice; at least one structural conduction signal sensing unit disposed within the housing assembly and mechanically coupled to a structural component receiving structural vibrations caused by the wearer's voice, for converting structural vibrations and / or structural conduction sound into structural conduction signals; a signal processing module connected to both the air conduction microphone and the structural conduction signal sensing unit; and a receiver connected to the signal processing module. The signal processing module is configured to: acquire the air conduction sound signals and the structural conduction signals; construct fusion input information based on the correspondence between the air conduction sound signals and the structural conduction signals; input the fusion input information to a speech processing model to obtain an output result for distinguishing between the wearer's voice and non-wearer's voice; reduce the proportion of the wearer's voice in the hearing aid output based on the output result, while maintaining or enhancing the non-wearer's voice, and output this output through the receiver.

[0023] This invention solves the problems of air conduction microphones having difficulty in stably distinguishing between the wearer's own voice and the voice of non-wearers, the loss of effective external sound when uniformly attenuating mixed inputs, and the unstable effect of existing self-voice suppression. Furthermore, it provides a mechanical coupling structure that does not rely on structural sensing units, with a single device type, fixed position, or installation posture. This allows the product to select the sensor type and installation position based on internal space, structural vibration modes, assembly process, and target frequency response, while maintaining effective extraction of the wearer's own voice. Attached Figure Description

[0024] Figure 1 This is a schematic diagram showing the installation of the structural conduction signal sensing unit in the hearing aid device based on the fusion of air conduction signals and structural conduction signals, according to an embodiment of the present invention. The structural conduction signal sensing unit is located between the internal support and the middle of the upper shell.

[0025] Figure 2 This is a schematic diagram of a hearing aid device based on the fusion of air conduction signals and structural conduction signals according to an embodiment of the present invention, in which the structural conduction signal sensing unit is located at the center of the internal support and is installed in a forward orientation.

[0026] Figure 3 This is a schematic diagram of a hearing aid device based on the fusion of air conduction signals and structural conduction signals according to an embodiment of the present invention, in which the structural conduction signal sensing unit is located at the center of the internal support and is installed horizontally.

[0027] Figure 4 This is a schematic diagram showing the installation of the structural conduction signal sensing unit in the hearing aid device based on the fusion of air conduction signals and structural conduction signals, according to an embodiment of the present invention. The structural conduction signal sensing unit is located between the internal support and the middle of the lower housing.

[0028] Figure 5 This is a schematic diagram showing the installation of the structural conduction signal sensing unit at the rear of the hearing aid device in an embodiment of the present invention, which is based on the fusion of air conduction signals and structural conduction signals.

[0029] Figure 6 This is a schematic diagram showing the installation of the structural conduction signal sensing unit between the internal support and the front of the lower housing in a hearing aid device based on the fusion of air conduction signals and structural conduction signals, according to an embodiment of the present invention.

[0030] Figure 7 This is a schematic diagram of the signal processing and neural network architecture in a hearing aid device based on the fusion of air conduction signals and structural conduction signals according to an embodiment of the present invention.

[0031] Figure 8 This diagram illustrates the input of air conduction signals and structural conduction signals, as well as the dual-channel output of the wearer's voice and the non-wearer's voice.

[0032] Figure 9 This is a flowchart of a preferred embodiment of the self-sound suppression method for hearing aids based on the fusion of air conduction signals and structural conduction signals according to an embodiment of the present invention.

[0033] Figure 10 This is a technical framework diagram of the self-sound suppression system of a hearing aid device based on the fusion of air conduction signals and structural conduction signals, according to an embodiment of the present invention.

[0034] Figure 11 A schematic diagram of a terminal provided in an embodiment of the present invention.

[0035] Explanation of reference numerals: 10—Air conduction microphone; 11—Pickup hole; 12—Structural conduction signal sensing unit; 20—Upper housing; 21—Pickup end; 22—Lower housing; 30—Battery; 40—Functional interface or control component; 50—Receiver; 60—Acoustic duct or ear output component; 70—Internal support. Detailed Implementation

[0036] To make the objectives, technical solutions, and effects of this invention clearer and more explicit, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0037] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content, operations, or steps, nor does it require execution in the described order. For example, some operations or steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation.

[0038] It should be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0039] It should be understood that, in order to clearly describe the technical solutions of the embodiments of the present invention, the terms "first" and "second" are used in the embodiments of the present invention to distinguish identical or similar items with essentially the same function and effect. For example, "first control information" and "second control information" are only used to distinguish different control information and do not limit their order.

[0040] Those skilled in the art will understand that the words "first" and "second" do not limit the quantity or the order of execution, and that the words "first" and "second" do not necessarily imply that they are different.

[0041] It should also be understood that the term “and / or” as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0042] To address the challenges of air-conduction microphones in reliably distinguishing between the wearer's own voice and non-wearer's voice, the loss of effective external sound due to uniform attenuation of mixed inputs, and the unstable effect of existing self-sound suppression, this invention provides a hearing aid device based on the fusion of air-conduction signals and structural conduction signals. In this embodiment, "self-sound" refers to at least one of the following: air-conduction sound, bone conduction sound, tissue conduction sound, ear canal sound, and device structural vibration, generated by the wearer's own voice. "Non-wearer's voice" refers to voice emitted by other individuals that primarily reaches the hearing aid device via air transmission.

[0043] Combination Figure 1 As shown, the hearing aid device includes: a housing assembly, at least one air conduction microphone 10, at least one structure conduction signal sensing unit 12, a signal processing module, and a receiver 50. The housing assembly includes an upper housing 20 and a lower housing 22. Furthermore, the hearing aid device of this embodiment also includes a pickup end 21, a battery 30, a functional interface or control component 40, an acoustic duct or ear output component 60, and an internal support 70, etc. In practical application scenarios, the above components can be added, removed, or combined depending on the type of hearing aid device.

[0044] Specifically, the air conduction microphone 10 in this embodiment is used to collect ambient sound, non-wearer's voice, and wearer's air conduction sound. The structural conduction signal sensing unit 12 is a sensing device capable of detecting mechanical vibrations or structural conduction sound transmitted through bones, soft tissues, ear contact structures, hearing aid housing, or internal support 70 via direct or indirect mechanical coupling, and outputting corresponding electrical signals. In practical applications, the structural conduction signal sensing unit 12 is disposed within the housing assembly and mechanically coupled to a structural component that receives structural vibrations caused by the wearer's voice, used to convert structural vibrations and / or structural conduction sound into structural conduction signals. A signal processing module is connected to both the air conduction microphone 10 and the structural conduction signal sensing unit 12. The upper housing 20 can be a single-piece molding or formed by connecting the upper housing 20 and the pickup end 21. The pickup hole 11 communicates with the air conduction microphone 10. The internal support 70 is used to support the battery 30, circuit board, sensor, receiver 50, or other components, and together with the upper housing 20 and lower housing 22, forms the internal structure of the device. The air-conducting microphone 10 can be a single microphone or N microphones. Multiple microphones can be used for beamforming, spatial noise reduction, or selecting at least one for correlation analysis with the structure-conducted signal.

[0045] In one implementation, the structural conduction signal sensing unit 12 of this embodiment includes any one or more of the following: a voice pick-up unit (VPU), a piezoelectric vibration sensor, a MEMS accelerometer, a contact microphone, a bone conduction sound sensor, an inertial sensor, a strain sensor, a force sensor, a vibration velocity sensor, or a displacement sensor. It may also include a base, a mechanical coupling layer, a connector, and a signal conditioning circuit. The aforementioned sensors can characterize the structural conduction signal by detecting changes in acceleration, velocity, displacement, strain, force, or contact sound pressure; the specific physical quantity sensed is not limited. The structural conduction signal sensing unit 12 is connected to the internal support 70, the upper housing 20, the lower housing 22, or other structural components via the base or mechanical coupling layer. The connection should allow the target acoustic vibration to effectively enter the structural sensing unit, while reducing direct airborne acoustic excitation through material damping, partial sealing, or structural isolation. The structural conduction signal sensing unit 12 is connected to the structural component by any one of the following methods: bonding, snap-fitting, fastening, pre-compression, elastic pad coupling, or integral molding, so that the structural vibration generated by the wearer's voice is transmitted to the structural conduction signal sensing unit 12, and the influence of external air conduction sound on the output of the structural conduction signal is reduced.

[0046] When the wearer speaks, the vocal cords vibrate and are transmitted to the hearing aid via the skull, tissues, auricle, or ear canal contact area, further forming structural vibrations along the shell and internal support 70. The structural conduction signal sensing unit 12 collects this structural vibration or structural conduction sound, thus its output has a stable correspondence with the start and end, fundamental frequency, harmonics, and envelope of the wearer's own voice. When a non-wearer speaks, their sound mainly reaches the air conduction microphone 10 via the air. Since the non-wearer's sound does not form an equivalent structural conduction path through the wearer's body, the corresponding response in the structural conduction signal sensing unit 12 is usually lower. The processing module can compare the energy ratio, coherence, and time delay of the structural conduction channel and the air conduction channel to avoid misinterpreting external speech as the wearer's own voice based solely on the air conduction sound level.

[0047] Furthermore, the structural conduction signal sensing unit 12 of this embodiment is disposed in at least one of the following preset installation areas: the middle part of the upper housing 20, the front part of the upper housing 20, the rear part of the upper housing 20, between the internal support 70 and the upper housing 20, between the internal support 70 and the lower housing 22, the center of the internal support 70, the middle part of the lower housing 22, the front part of the lower housing 22, the rear part of the lower housing 22, the rear part of the hearing aid, the pickup end 21, the end of the receiver 50, the inner wall of the upper housing 20, the inner wall of the lower housing 22, or an area on the internal support 70 where a mechanical coupling path can be formed. Moreover, the installation posture, sensing direction, and base orientation of the structural conduction signal sensing unit 12 are set according to the housing space and the direction of structural vibration transmission.

[0048] Specifically, Figure 2 An installation method for the structural conductive signal sensing unit 12 is shown; in this illustrated embodiment, the structural conductive signal sensing unit 12 employs a VPU assembly. Its base can face the main bearing surface of the support, aligning the sensor's sensing direction with the predetermined structural vibration direction. This design facilitates the use of the overall vibration transmitted by the support while maintaining a certain structural isolation from external airborne noise.

[0049] Figure 3 The structural conductive signal sensing unit 12 is shown located in the central region of the internal support 70 and is mounted laterally. Lateral mounting can reduce occupancy when internal height is limited, or it can align the sensor's sensitive direction with another dominant vibration direction. The sensor base should form a complete support and stable preload to reduce additional resonance caused by assembly looseness.

[0050] Figure 4 The structural conduction signal sensing unit 12 is shown to be located between the inner support 70 and the middle of the lower housing 22. This area can utilize structural vibrations obtained from the contact between the lower housing 22 and the wearer's ear or head, and the coupling stiffness can be adjusted by elastic pads, adhesive layers, or pre-compression structures.

[0051] Figure 5The structural conduction signal sensing unit 12 is shown to be located at the tail of the hearing aid. When the tail has a structural path connected to the upper housing 20, the lower housing 22, or the internal support 70, it can be used to acquire the self-sound vibrations propagating along the housing; the processing module can recalibrate the calibration parameters or speech processing model parameters between the air conduction channel and the structural conduction channel for this location.

[0052] Figure 6 The structural conductive signal sensing unit 12 is shown to be located between the internal support 70 and the front of the lower housing 22. When this position is close to the pickup end 21 or the wearing contact area, a higher structural conductive signal-to-noise ratio may be obtained; crosstalk during the operation of the air-conducting microphone 10 can be reduced through a mechanical isolation structure.

[0053] It should be noted that the installation position of the structural conduction signal sensing unit 12 is not limited to the position shown in the attached drawings. Any location that can establish an effective mechanical coupling path from the wearer's vocal structure vibration to the structural sensing unit, and that enables the resulting structural conduction signal to have self-sound discrimination capability, can be used as the installation area. In other embodiments, the structural conduction signal sensing unit 12 may also be located between the internal support 70 and the upper housing 20, on the inner wall of the upper housing 20, on the inner wall of the lower housing 22, near the pickup end 21, near the receiver 50, near the support structure of the battery 30, in the circuit board fixing area, in the acoustic duct fixing area, or in other locations within the housing. The above locations are examples, not exhaustive; the criterion is whether a mechanical coupling path that meets the requirements of self-sound discrimination and fusion processing can be formed. The structural conduction signal sensing unit 12 can be installed upright, horizontally, vertically, tilted, or at different angles relative to the support with its base. Furthermore, the structural conduction signal sensing unit 12 can be rigidly connected to a structural component or connected via an intermediate layer with predetermined stiffness and damping. Rigid connections are beneficial for improving the transmission of high-frequency vibrations; elastic or damped connections can adjust resonance peaks and reduce impact. Connection parameters can be designed in conjunction with digital equalization. Specific locations, quantities, and orientations can be determined through structural modal analysis, prototype measurements, or online calibration.

[0054] In other implementations, when multiple structural conduction signal sensing units 12 are used, they can be arranged on the upper shell 20 side, the lower shell 22 side, the center of the support, or the tail. The processing module can select the sensing unit with the highest current structural conduction signal-to-noise ratio, or perform weighted, differential, array processing, or consistency detection on multiple structural conduction signal sensing units 12 to improve stability under different wearing conditions. The structural conduction signal sensing unit 12 does not need to be completely unresponsive to airborne sound, nor is it required that its output only contains self-sound. As long as, under predetermined usage conditions, the structural sensing unit has a sufficient discriminative signal-to-noise ratio for structural vibrations or structural conduction sound caused by the wearer's voice, relative to the non-wearer's airborne speech, the processing module can achieve stable recognition by combining multi-feature and confidence gating.

[0055] Furthermore, since the air-conducting microphone 10 and the structure sensing unit correspond to different physical conduction paths, when the wearer speaks, the two signals form a learnable or calibrable correspondence in terms of time variation, energy, spectrum, envelope, fundamental frequency, harmonics, and channel correlation. Therefore, the signal processing module can construct fused input information based on the correspondence between the air-conducting sound signal and the structure-conducting signal. Non-wearer speech mainly reaches the microphone through air conduction, and typically lacks a matching response in the structure-conducting signal. The processing module thus obtains speech source discrimination information that differs from information obtained using only the air-conducting microphone 10.

[0056] The signal processing module in this embodiment is configured to: acquire airborne sound signals and structural sound signals; construct fused input information based on the correspondence between the airborne sound signals and the structural sound signals; input the fused input information into a speech processing model to obtain an output result for distinguishing between the wearer's speech and the non-wearer's speech. This output result can be two signals: one for the wearer's speech estimation and one for the non-wearer's speech estimation; another can be a combination of one speech estimation signal and the wearer's or self-speech confidence score; and yet another can be a time-domain signal, time-frequency representation, mask, gain, category label, category probability, or other information used to control the two types of speech components respectively. Then, based on the output result, the proportion of the wearer's speech in the hearing aid output is reduced, while the non-wearer's speech is maintained or enhanced, and then output through the receiver 50. Thus, the system can perform different hearing aid processing for the two speech sources, rather than uniformly attenuating the mixed input.

[0057] In practical applications, the processing module can first perform one or more of the following processing on the airborne sound signal and the structure-conducted sound signal: sampling rate unification, time synchronization, DC removal, filtering, gain normalization, and channel calibration. Then, the two time-domain signals can be used directly, or a time-frequency transformation can be performed on the two signals to obtain an input representation suitable for use by the speech processing model.

[0058] The processing module can fuse airborne acoustic signals and structurally conducted acoustic signals in the time or frequency domain to obtain fused input information. The fused input information includes at least one of the following: the time-domain waveform, time-frequency representation, amplitude, phase, band energy, envelope, fundamental frequency, harmonics, channel amplitude difference, channel amplitude ratio, cross-correlation, coherence, time delay, speech activity state, and self-speech confidence of the airborne acoustic signal and the structurally conducted acoustic signal. The frequency band can be the entire frequency band or at least one sub-frequency band. The above features can be calculated within the entire frequency band or at least one sub-frequency band and can be spliced, weighted, mapped, or encoded to form the input of the speech processing model. This fusion is not limited to a specific frequency point, fixed frequency band, fixed sampling parameters, or fixed model structure. Specifically, in the time-frequency unit (f,n), f is the frequency point index and n is the time frame index. The i-th airborne sound signal can be represented as x_i(f,n)=e_i(f,n)+s_i(f,n)+d_i(f,n), where e_i represents ambient sound and non-wearer speech, s_i represents the wearer's airborne self-sound component, and d_i represents noise. The structurally conducted signal can be represented as v(f,n)=q(f,n)+η(f,n), where q represents structural vibration or structurally conducted sound caused by the wearer's voice, and η represents non-target vibration and sensor noise. Since s_i and q originate from the same wearer's voice event, there is an estimable temporal, energy, and spectral correspondence between them. However, e_i typically does not have the same structurally conducted response. Therefore, this physical difference forms the basis for the joint identification and separation of the wearer's speech using airborne and structurally conducted signals.

[0059] When speech is detected in the airborne acoustic signal and a response corresponding to its time variation or acoustic characteristics exists in the structurally conducted signal, it can be determined that the current speech mainly originates from the wearer, or the self-sound confidence level can be increased. When speech is present in the airborne acoustic signal but no corresponding response is formed in the structurally conducted signal, it can be determined that the current speech mainly originates from a non-wearer. For areas without speech, non-target vibrations, collisions, or changes in wearing status, the processing module can reduce false positives through thresholds, state machines, classification models, or multi-frame smoothing. The self-sound determination result or self-sound confidence level in this embodiment can directly control the gain of the wearer's speech or serve as an auxiliary input to the speech processing model. In a preferred embodiment, the airborne acoustic signal, the structurally conducted signal, or their time-frequency representation can be jointly input into the speech processing model, allowing the model to utilize the complementary information between the two continuous signals, rather than simply treating the structural sensing unit as a binary switch.

[0060] The speech processing model in this embodiment can employ a rule-based decision model, a statistical classification model, an adaptive filtering model, a blind source separation model, a priori source separation model, a neural network model, or a combination thereof. The model can output time-frequency masks, time-frequency representations, or time-domain waveforms corresponding to the wearer's speech and non-wearer's speech. It can also output wearer's self-voice estimation, and the non-wearer's speech obtained by removing the self-voice estimation from the mixed signal.

[0061] In one implementation, the speech processing model outputs a wearer's speech branch and a non-wearer's speech branch based on the joint input of the airborne and structured acoustic signals. In another implementation, the model first estimates the wearer's self-voice component in the airborne acoustic signal based on the structured acoustic signal, and then subtracts or attenuates the self-voice component from the airborne acoustic signal. In yet another implementation, self-voice confidence is used to control frequency-dependent or time-varying gain. The above methods can be used individually or in combination.

[0062] The neural network model in this embodiment includes an input layer, a temporal modeling layer, and an output layer. The input layer receives an input representation obtained from at least one airborne acoustic signal and at least one structural acoustic signal. The temporal modeling layer includes a Long Short-Term Memory (LSTM) network, or any one or more of recurrent neural networks, gated recurrent units, temporal convolutional networks, convolutional neural networks, Transformer networks, and state-space networks. The output layer employs at least one of the following output methods: outputting two signals, one for wearer's speech estimation and one for non-wearer's speech estimation; outputting one speech estimation signal and the wearer's speech activity detection result or self-speech confidence; outputting the wearer's speech estimation and obtaining the non-wearer's speech estimation from the airborne acoustic signal based on the wearer's speech estimation; and outputting a time-frequency mask, gain, category label, or category probability corresponding to the wearer's speech and non-wearer's speech to distinguish and control the wearer's speech and non-wearer's speech respectively.

[0063] Specifically, such as Figure 7 As shown, in one specific implementation, the speech processing model employs an LSTM (Long Short-Term Memory) network. The network input includes fused input information obtained from at least one airborne acoustic signal and at least one structurally conducted signal. When multiple airborne microphones 10 or multiple structurally conducted signal sensing units 12 are provided, at least one of them can be selected, or multiple signals and their features can be concatenated, weighted, or encoded before being sent to the network.

[0064] Airborne sound signals and structurally conducted sound signals can be time-synchronized, framed, normalized, and optionally time-frequency transformed to form frame-level input features. The fusion feature of the nth frame can be represented as Z(n) = Concat(F_AIR(n), F_STRUCT(n), F_cross(n)), where F_AIR(n) and F_STRUCT(n) represent airborne and structurally conducted features, respectively, and F_cross(n) represents optional cross-channel features such as amplitude difference, amplitude ratio, correlation, coherence, time delay, or self-sound confidence. The temporal modeling layer performs temporal modeling based on the current frame fusion feature Z(n) and the previous state h(n-1) to obtain the current state h(n). The processing relationship can be summarized as h(n) = LSTM(Z(n), h(n-1)). LSTM can be single-layer or multi-layer, unidirectional or bidirectional; in real-time hearing aid processing, a unidirectional streaming LSTM with cached historical states is preferred to reduce algorithm latency and computational load.

[0065] like Figure 8 As shown, in a specific implementation of the dual-channel output, the network input includes at least one airborne acoustic signal (i.e., self-sound signal) and one structurally conducted signal (i.e., non-self-sound signal). The airborne acoustic signal may include the wearer's self-sound, non-wearer's speech, and environmental noise, while the structurally conducted signal primarily provides a reference related to the wearer's vocalization. After fusion processing of the airborne and structurally conducted signals and modeling using a speech processing model, the first output is the wearer's speech estimate y_wearer (i.e., the corresponding...). Figure 8 The second output is the non-wearer speech estimate y_non_wearer (i.e., the corresponding mic signal in the microphone), and the third output is the non-wearer speech estimate y_non_wearer (i.e., the corresponding microphone signal in the microphone). Figure 8 (VPU signal in the image). The two outputs can be time-domain waveforms, time-frequency representations, time-frequency masks applied to the reference signal, or gain coefficients, and can be used for gain adjustment, hearing compensation, or subsequent service processing, respectively. The first and second outputs mentioned above represent two distinguishable output paths, and their interface order can be interchanged, which does not constitute a restriction on the arrangement of the output ports.

[0066] In the second output method, the network output layer can output a speech signal and the wearer's voice activity detection (VAD) result or self-speech confidence. For example, the network can output a non-wearer's speech estimate and use the wearer's VAD to control the self-speech suppression intensity; or it can output the wearer's speech estimate and the wearer's VAD, and then subtract or attenuate the wearer's speech estimate from the airborne sound signal to obtain a hearing aid signal that preserves the non-wearer's speech.

[0067] In other output methods, the network output layer can output wearer or non-wearer category labels, category probabilities, speaking state, time-frequency mask, frequency band gain, source presence probability, or other control information that can distinguish the source of speech. As long as the output can be used to determine whether the current speech belongs to a wearer or a non-wearer, or can perform different gain controls, estimations, or separations for the two types of speech, the self-speech suppression function of this invention can be achieved. The LSTM described above is a specific example and does not limit the speech processing model to using an LSTM structure.

[0068] When the speech processing model employs a training-required model, training data containing airborne sound signals, structurally conducted sound signals, wearer's speech labels, non-wearer's speech labels, and environmental noise can be used to enable the model to learn the correspondence between the two physical conduction paths and the speech source. Training objectives may include at least one of the following: classification of wearer's and non-wearer's speech, time-domain or frequency-domain reconstruction, mask estimation, source separation, speech intelligibility, and output consistency. Specific loss functions, time-frequency parameters, and model structures can be selected based on device computing power, target latency, and sound quality requirements.

[0069] The training data for the speech processing model can cover one or more of the following scenarios: only wearer's speech, only non-wearer's speech, alternating speech between wearer and non-wearer, overlapping speech between the two, no speech, environmental noise, device collision, changes in wearing status, and structural sensing units in different coupling states, to improve the model's adaptability in real-world hearing aid scenarios. Training data for airborne sound signals can be obtained from actual acquired signals or simulated mixed signals. Training data for structurally conducted sound signals can be obtained from signals acquired through actual mechanical vibration, bone conduction, or tissue conduction, or from reference signals generated based on the device's transmission characteristics. During training, signal sound level, frequency response, time delay, noise, reverberation, and sensor sensitivity can be perturbed. For hearing aids with limited computing power and low latency requirements, a combination of lightweight networks, frame-by-frame streaming processing, state caching, model quantization, model pruning, or rule-based decision-making with neural networks can be used.

[0070] Furthermore, the speech processing model obtains the wearer's speech estimate y_wearer and the non-wearer's speech estimate y_non_wearer based on the fused input of the airborne and structurally conducted signals. These two estimates can correspond to time-domain signals, time-frequency signals, or gain control values. When one or more non-wearer speakers are present, their speech can be categorized under the non-wearer speech estimate. When the model outputs two time-frequency representations, the processing module can recover the time-domain signal through inverse time-frequency transformation or frame-level synthesis; when the model directly outputs a time-domain waveform, it can directly proceed to subsequent hearing aid processing; when the model outputs a mask or gain, it can be applied to the airborne sound signal, reference signal, or corresponding intermediate representation.

[0071] The final hearing aid output can be expressed as y = α·y_non_wearer + β·y_wearer, where α is the non-wearer's speech gain coefficient and β is the wearer's self-voice preservation coefficient. α and β are preset fixed values ​​or dynamically adjusted values, respectively. The dynamic adjustment value is determined based on at least one of the following: self-voice confidence level, user self-voice suppression level, input sound level, hearing compensation parameters, wearing status, and usage scenario. When α increases, the proportion of the non-wearer's speech in the hearing aid output is increased; when β decreases, the wearer's self-voice is further suppressed; when β increases, more natural vocal feedback is preserved. α and β can be set to fixed values ​​or dynamically adjusted values, respectively. This expression is used to illustrate that the two speech streams can be adjusted independently and does not constitute a limitation on the fusion method.

[0072] In addition to proportional fusion, this embodiment can also reduce the wearer's self-voice by canceling the wearer's self-voice estimation from the airborne sound signal, reducing the frequency-dependent gain within the self-voice interval, performing dynamic range compression on the wearer's speech branch, or generating a occlusion effect compensation signal. The self-voice suppression intensity can be determined based on at least one of the following: self-voice confidence level, input sound level, user audiogram, wearing seal, wearing status, user settings, and current usage scenario. The system can set the attack time and release time to smooth the suppression intensity, avoiding abrupt changes when switching between the wearer's and non-wearer's speech. The non-wearer's speech estimation can be further processed by beamforming, noise reduction, equalization, dynamic range compression, and hearing compensation. The fused hearing aid signal can be processed by gain limiting, amplitude limiting, and anti-feedback processing. The above subsequent processing does not change the core processing relationship of jointly identifying and selectively suppressing the wearer's self-voice using airborne and structurally conducted signals. Therefore, the “self-sound suppression” referred to in this embodiment is not limited to completely eliminating the wearer’s voice, but includes one or more processing methods that reduce the absolute energy or relative proportion of the wearer’s voice in the final hearing aid output while maintaining the non-wearer’s voice and effective ambient sound.

[0073] It should be noted that in this embodiment, the air conduction microphone 10, the structure conduction signal sensing unit 12, the processing module, and the receiver 50 can be integrated into the same housing, or partially disposed in different components connected by wired or wireless means. The structure sensing unit can output analog signals, digital signals, single-axis signals, or multi-axis signals. The voice processing model can be a user-specific model, a device model-specific model, or a general model, and the correspondence between the air conduction signal and the structure conduction signal can be initialized or calibrated through user reading, factory testing, or online reliability testing.

[0074] Based on the above embodiments, the present invention also provides a self-sound suppression method for hearing aids based on the fusion of air conduction signals and structural conduction signals. The method of this embodiment can be applied to the hearing aids described above, and also to terminal devices such as computers. Specifically, as... Figure 9 As shown in the figure, the self-sound suppression method of the hearing aid device based on the fusion of air conduction signals and structural conduction signals in this embodiment includes the following steps: Step S100: Simultaneously acquire at least one airborne acoustic signal and at least one structural acoustic signal; Step S200: Construct fused input information based on the correspondence between the air-conducted sound signal and the structure-conducted sound signal; Step S300: Input the fused input information into the speech processing model to obtain the output result used to distinguish between the wearer's speech and the non-wearer's speech; Step S400: Based on the output result, reduce the proportion of the wearer's voice in the hearing aid output, while maintaining or enhancing the non-wearer's voice, generate a self-sound-suppressed hearing aid signal, and output it through the receiver.

[0075] The speech processing model in this embodiment obtains wearer speech estimation and non-wearer speech estimation through at least one of the following methods: outputting time-frequency masks corresponding to wearer speech and non-wearer speech respectively and applying them to the reference signal; directly mapping to obtain two time-frequency representations; directly generating two time-domain signals; estimating the wearer's self-voice component in the air-conducted sound signal based on the structure-conducted signal, and removing all or part of the wearer's self-voice component from the air-conducted sound signal.

[0076] The speech processing model in this embodiment is trained using training samples that include wearer speech, non-wearer speech, environmental noise, and corresponding structural conduction signals. The training objective is to constrain at least one of the following for wearer speech and non-wearer speech: source classification, speech activity detection, channel attribution, separation, speech intelligibility, and waveform fidelity.

[0077] When reducing the proportion of the wearer's voice in the hearing aid output based on the output results, while maintaining or enhancing the non-wearer's voice, this embodiment increases the self-voice confidence and reduces the proportion of the wearer's voice component in the hearing aid output when the correspondence between the structurally conducted signal and the airborne sound signal meets the wearer's self-voice determination condition. When voice is present in the airborne sound signal but the structurally conducted signal does not form a corresponding response, it is determined to be non-wearer's voice, and the output of the non-wearer's voice component is maintained or enhanced.

[0078] The implementation principle of each step in the self-sound suppression method of the hearing aid device based on the fusion of air conduction signal and structure conduction signal in this embodiment is the same as the functional principle of each module in the hearing aid device embodiment, and will not be elaborated further here.

[0079] Based on the above embodiments, this embodiment also provides a self-sound suppression system for hearing aids based on the fusion of air conduction signals and structural conduction signals. This system is used to implement the steps in the above method embodiments. Figure 10 As shown, the system includes: a conductive signal acquisition module, a fusion input construction module, a speech separation processing module, and an output control module. Specifically, the conductive signal acquisition module is used to simultaneously acquire at least one airborne acoustic signal and at least one structural acoustic signal. The fusion input construction module is used to construct fusion input information based on the correspondence between the airborne acoustic signal and the structural acoustic signal. The speech separation processing module is used to input the fusion input information into a speech processing model to obtain an output result that distinguishes between the wearer's speech and the non-wearer's speech. The output control module is used to reduce the proportion of the wearer's speech in the hearing aid output based on the output result, while maintaining or enhancing the non-wearer's speech, generating a self-suppressed hearing aid signal, and outputting it through a receiver.

[0080] The principle of each module in the self-sound suppression system of the hearing aid device based on the fusion of air conduction signals and structural conduction signals in this embodiment is the same as the functional principle of each module in the hearing aid device embodiment, and will not be elaborated further here.

[0081] The above embodiments are illustrative and not limiting of the invention. Any method that simultaneously utilizes airborne sound signals and structurally conducted signals acquired by a mechanically coupled structural sensing unit to identify, estimate, or separate the wearer's and non-wearer's speech in a hearing aid device, and selectively reduces the proportion of the wearer's speech in the hearing aid output, falls within the scope defined by the corresponding claims.

[0082] Compared to existing technologies that rely solely on air-conducting microphones, this invention leverages the complementarity between air-conducting and structural-conducting inputs to provide an independent physical reference for determining whether sound originates from the wearer. This improves the stability of wearer self-voice recognition and reduces the probability of misinterpreting non-wearer speech as self-voice. Instead of solely using a structural sensing unit to generate a self-voice switching signal, this invention uses both air-conducting and structural-conducting signals for estimating wearer and non-wearer speech, creating independently controllable processing pathways for both types of speech. This reduces wearer self-voice energy while maintaining the clarity and continuity of external communication speech. The technical solution of this invention includes complementary structural and algorithmic solutions. Structurally, the hearing aid device simultaneously incorporates at least one air-conducting microphone and at least one structural sensing unit mechanically coupled to the device structure, forming two physical input paths: air-conducting and structural-conducting. Algorithmically, at least one air-conducting signal and at least one structural-conducting signal are jointly processed to output a signal or control information that distinguishes between wearer and non-wearer speech, selectively suppressing wearer self-voice accordingly. This invention does not limit the specific fusion method, frequency band range, network structure, or output format.

[0083] This invention uses the effective mechanical coupling area as the overarching concept for the installation position of the structural conduction signal sensing unit. The center of the bracket, the area between the bracket and the upper and lower shells, the front or middle part of the lower shell, and the tail part in the attached drawings are all optional embodiments and will not lose their applicability due to product internal space or structural iterations.

[0084] Based on the above embodiments, the present invention also provides a terminal, the principle block diagram of which can be as follows: Figure 11 As shown. The terminal may include one or more processors 100 ( Figure 11 (Only one is shown in the image), memory 101, and computer program 102 stored in memory 101 and executable on one or more processors 100. For example, a self-sound suppression program for a hearing aid based on the fusion of air-conducted and structure-conducted signals. When one or more processors 100 execute computer program 102, they can implement the various steps in the self-sound suppression method embodiment for a hearing aid based on the fusion of air-conducted and structure-conducted signals. Alternatively, when one or more processors 100 execute computer program 102, they can implement the functions of various modules / units in the self-sound suppression device embodiment for a hearing aid based on the fusion of air-conducted and structure-conducted signals, which is not limited here.

[0085] In one embodiment, the processor 100 may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0086] In one embodiment, memory 101 can be an internal storage unit of the terminal, such as a hard disk or RAM. Memory 101 can also be an external storage terminal of the terminal, such as a plug-in hard disk, smart media card (SM), secure digital card (SD), flash card, etc., all equipped on the terminal. Furthermore, memory 101 can include both internal and external storage units. Memory 101 is used to store computer programs and other programs and data required by the terminal. Memory 101 can also be used to temporarily store data that has been output or will be output.

[0087] Those skilled in the art will understand that Figure 11 The block diagram shown is merely a partial structural diagram related to the present invention and does not constitute a limitation on the terminal to which the present invention is applied. A specific terminal may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0088] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), direct memory bus RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0089] 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 of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A hearing aid device based on the fusion of air conduction signals and structural conduction signals, characterized in that, The hearing aid device includes: Housing assembly; At least one air-conducting microphone is used to acquire air-conducted sound signals containing the wearer's own voice and non-wearer's voice; At least one structural conduction signal sensing unit is disposed within the housing assembly and mechanically coupled to a structural component that receives structural vibrations caused by the wearer's voice, for converting structural vibrations and / or structural conduction sound into structural conduction signals; The signal processing module is connected to both the air conduction microphone and the structure conduction signal sensing unit. The receiver is connected to the signal processing module; The signal processing module is configured as follows: Acquire the air-conducted sound signal and the structure-conducted sound signal; The fused input information is constructed based on the correspondence between the air-conducted sound signal and the structure-conducted sound signal; The fused input information is input into the speech processing model to obtain an output result used to distinguish between the wearer's speech and the non-wearer's speech; Based on the output results, the proportion of the wearer's voice in the hearing aid output is reduced, while the non-wearer's voice is maintained or enhanced, and then output through the receiver. The speech processing model includes any one or more of the following: rule-based decision model, statistical classification model, adaptive filtering model, source separation model, and neural network model; the speech processing model is used to determine the wearer's self-voice interval or self-voice confidence based on the fusion input information of the air-conducted signal and the structure-conducted signal, and to estimate the wearer's speech component and non-wearer's speech component in the air-conducted sound signal.

2. The hearing aid device based on the fusion of air conduction signals and structural conduction signals according to claim 1, characterized in that, The structural conductive signal sensing unit includes any one or more of the following: a vibration pickup unit, a piezoelectric vibration sensor, a MEMS accelerometer, a contact microphone, a bone conduction pickup sensor, an inertial sensor, a strain sensor, a force sensor, a vibration velocity sensor, or a displacement sensor.

3. The hearing aid device based on the fusion of air conduction signals and structural conduction signals according to claim 2, characterized in that, The structural conductive signal sensing unit is connected to the structural component through any one of the following methods: bonding, snap-fitting, fastening, pre-compression, elastic pad coupling, or integral molding, so that the structural vibration generated by the wearer's voice is transmitted to the structural conductive signal sensing unit, and the influence of external airborne sound on the output of the structural conductive signal is reduced.

4. The hearing aid device based on the fusion of air conduction signals and structural conduction signals according to claim 1, characterized in that, The structural conductive signal sensing unit is disposed in at least one of the following preset installation areas: the middle part of the upper housing, the front part of the upper housing, the rear part of the upper housing, between the internal support and the upper housing, between the internal support and the lower housing, the center of the internal support, the middle part of the lower housing, the front part of the lower housing, the rear part of the lower housing, the rear part of the hearing aid, the pickup end, the end of the receiver, the inner wall of the upper housing, the inner wall of the lower housing, or an area on the internal support where a mechanical coupling path can be formed; The installation posture, sensitive direction, and base orientation of the structural conductive signal sensing unit are set according to the housing space and the direction of structural vibration transmission.

5. The hearing aid device based on the fusion of air conduction signals and structural conduction signals according to claim 1, characterized in that, The fused input information includes at least one of the following: the time-domain waveform, time-frequency representation, amplitude, phase, frequency band energy, envelope, fundamental frequency, harmonics, channel amplitude difference, channel amplitude ratio, cross-correlation, coherence, time delay, speech activity state, and self-sound confidence of the air-conducted sound signal and the structure-conducted sound signal.

6. The hearing aid device based on the fusion of air conduction signals and structural conduction signals according to claim 1, characterized in that, The neural network model includes an input layer, a temporal modeling layer, and an output layer; the input layer is used to receive an input representation obtained from at least one airborne acoustic signal and at least one structural acoustic signal; the temporal modeling layer includes a Long Short-Term Memory (LSTM) network, or any one or more of the following: a recurrent neural network, a gated recurrent unit, a temporal convolutional network, a convolutional neural network, a Transformer network, and a state-space network.

7. The hearing aid device based on the fusion of air conduction signals and structural conduction signals according to claim 6, characterized in that, The output layer employs at least one of the following output methods: Output two signals: wearer speech estimation and non-wearer speech estimation. Output one channel of speech estimation and wearer's speech activity detection results or self-speech confidence; Output the wearer's speech estimate, and obtain the non-wearer's speech estimate from the airborne sound signal based on the wearer's speech estimate; The output corresponds to the time-frequency mask, gain, category label, or category probability of the wearer's speech and the non-wearer's speech, in order to distinguish and control the wearer's speech and the non-wearer's speech separately.

8. The hearing aid device based on the fusion of air conduction signals and structural conduction signals according to claim 7, characterized in that, The signal processing module generates the hearing aid output according to y = α·y_non_wearer + β·y_wearer, where y_non_wearer is the speech estimation of the non-wearer, y_wearer is the speech estimation of the wearer, α is the speech gain coefficient of the non-wearer, and β is the self-voice preservation coefficient of the wearer; α and β are preset fixed values ​​or dynamically adjusted values, respectively. The dynamically adjusted value is determined based on at least one of the following: self-voice confidence, user self-voice suppression level, input sound level, hearing compensation parameters, wearing status, and usage scenario.

9. A method for self-sound suppression in hearing aids based on the fusion of air conduction signals and structural conduction signals, characterized in that, The self-sound suppression method is implemented using the hearing aid device based on the fusion of air conduction signals and structural conduction signals as described in any one of claims 1-8, and the method includes: Simultaneously acquire at least one airborne acoustic signal and at least one structural acoustic signal; The fused input information is constructed based on the correspondence between the air-conducted sound signal and the structure-conducted sound signal; The fused input information is input into the speech processing model to obtain an output result used to distinguish between the wearer's speech and the non-wearer's speech; Based on the output results, the proportion of the wearer's voice in the hearing aid output is reduced, while the non-wearer's voice is maintained or enhanced, generating a self-sound-suppressed hearing aid signal, which is then output through the receiver.

10. The self-sound suppression method for hearing aids based on the fusion of air conduction signals and structural conduction signals according to claim 9, characterized in that, The speech processing model obtains wearer speech estimation and non-wearer speech estimation through at least one of the following methods: The output time-frequency masks corresponding to the wearer's voice and the non-wearer's voice are respectively applied to the reference signal; Direct mapping yields two time-frequency representations; Directly generate two time-domain signals; The wearer's own voice component in the airborne sound signal is estimated based on the structure-conducted signal, and all or part of the wearer's own voice component is removed from the airborne sound signal.

11. The self-sound suppression method for hearing aids based on the fusion of air conduction signals and structural conduction signals according to claim 9, characterized in that, Based on the output results, reduce the proportion of the wearer's voice in the hearing aid output while maintaining or enhancing the non-wearer's voice, including: When the correspondence between the structurally conducted signal and the air-conducted sound signal satisfies the wearer's self-voice determination condition, the self-voice confidence is increased, and the proportion of the wearer's speech component in the hearing aid output is reduced. When the airborne sound signal contains speech but the structural sound signal does not generate a corresponding response, it is determined to be the speech of a non-wearer, and the output of the non-wearer speech component is maintained or enhanced.

12. The self-sound suppression method for hearing aids based on the fusion of air conduction signals and structural conduction signals according to claim 9, characterized in that, The speech processing model is trained using training samples that include wearer speech, non-wearer speech, environmental noise, and corresponding structural conduction signals. The training objective is to constrain at least one of the following for wearer speech and non-wearer speech: source classification, speech activity detection, channel attribution, separation, speech intelligibility, and waveform fidelity.

13. A self-sound suppression system for a hearing aid based on the fusion of air conduction signals and structural conduction signals, characterized in that, The system is used to implement the self-sound suppression method of the hearing aid device based on the fusion of air conduction signals and structural conduction signals as described in any one of claims 9-12, the method comprising: A conductive signal acquisition module is used to simultaneously acquire at least one airborne acoustic signal and at least one structural conductive signal; A fusion input construction module is used to construct fusion input information based on the correspondence between the air-conducted sound signal and the structure-conducted sound signal; The speech separation processing module is used to input the fused input information into the speech processing model to obtain an output result that distinguishes the wearer's speech from the non-wearer's speech; The output control module is used to reduce the proportion of the wearer's voice in the hearing aid output according to the output result, while maintaining or enhancing the non-wearer's voice, generating a self-sound-suppressed hearing aid signal, and outputting it through the receiver.

14. A terminal, characterized in that, The terminal includes a memory, a processor, and a self-sound suppression program for a hearing aid device based on the fusion of air conduction signals and structural conduction signals, stored in the memory and executable on the processor. When the processor executes the self-sound suppression program for a hearing aid device based on the fusion of air conduction signals and structural conduction signals, it implements the steps of the self-sound suppression method for a hearing aid device based on the fusion of air conduction signals and structural conduction signals as described in any one of claims 9-12.

15. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a self-sound suppression program for a hearing aid based on the fusion of air conduction signals and structural conduction signals. The self-sound suppression program for a hearing aid based on the fusion of air conduction signals and structural conduction signals implements the steps of the self-sound suppression method for a hearing aid based on the fusion of air conduction signals and structural conduction signals as described in any one of claims 9-12 on the computer-readable storage medium.

Images