Ear canal sound detection and analysis system

Abstract

The present application relates to the field of medical electronic equipment and biomedical signal processing technology, and particularly relates to an ear canal sound detection and analysis system, comprising a left and right ear back host, a pair of open earplugs and probe hoses, a pair of probe microphones and reference microphones, a binaural synchronization module, a pathological symmetry defect detection module, a trigger recording module and a pathological typing identification module, the system obtains an ear canal differential degree by differentiating binaural probe signals and an environment differential degree by differentiating binaural reference signals, and only when the ear canal differential degree exceeds a preset pathological threshold and the environment differential degree is lower than a preset environment threshold, a tinnitus candidate audio segment is started to be intercepted and stored, and then time-frequency features are extracted and sound source side and pathological typing results are output, the present application realizes 24h long-range capture and pathological typing of objective tinnitus without closing the external ear canal, and takes into account natural hearing reservation, hardware-level event filtering and system-level closed-loop self-optimization.

Description

Technical Field

[0001] This invention relates to the fields of medical electronic equipment and biomedical signal processing technology, specifically to a wearable ear canal sound detection and analysis system, which can be applied to long-term monitoring, automatic interception, and pathological classification of objective tinnitus. Background Technology

[0002] Tinnitus, as one of the most common pathological symptoms of the auditory nervous system, affects a very wide range of people. Epidemiological surveys indicate that the prevalence of tinnitus in adults is approximately 14% to 15%, with about 1% to 2% of patients experiencing objective tinnitus. Unlike subjective tinnitus, which is perceived solely by the patient, objective tinnitus usually originates from real physical sound sources such as blood turbulence in the periauricular vessels, muscle spasms (such as palatine myoclonus or stapedius muscle spasm), or abnormal opening of the Eustachian tube. The resulting sound pressure fluctuations can be perceived not only by the patient but also detected by clinicians using stethoscopes or acoustic sensors. Although objective tinnitus accounts for a small percentage of the overall tinnitus population, it often corresponds to a clear organic lesion. If the sound source can be accurately identified and located, and the pathological type identified, the clinical cure rate is far higher than that of subjective tinnitus. Therefore, objective detection of objective tinnitus has irreplaceable clinical value.

[0003] For subjective tinnitus, the psychoacoustic matching approach is currently widely used in clinical practice. European patent application EP2942010B1 discloses a Tinnitus diagnosis and test device. This system uses a fully digital audio and tone signal synthesizer to simultaneously generate synthesized sounds of various pitches, melodies, rhythms, and loudnesses for both ears, or interactively. A control module allows the patient to subjectively match the sounds they hear, thereby inferring the frequency, loudness, and acoustic characteristics of their tinnitus. However, this approach is only suitable for reconstructing subjective tinnitus. Its working mechanism relies on active sound stimulation and the patient's subjective feedback, and it cannot passively capture the actual physical sound source of objective tinnitus. Especially for objective tinnitus with intermittent and unilateral characteristics, any subjective matching results obtained when the patient is not in an attack phase lack physical evidence to support them.

[0004] In the direct measurement of the ear canal sound field, existing technologies have developed earbud-type acquisition devices integrating multiple microphones. US Patent Application Publication No. US11057701B2 discloses a method and device for in earcanal echo suppression. This system integrates an ambient sound microphone (ASM) to monitor external ambient sound and an ear canal microphone (ECM) to monitor the closed sound field within the ear canal. It uses a least mean square adaptive filter to differentially suppress the two signals, eliminating echo feedback generated by the wearer's own voice within the ear canal. However, this approach only uses dual microphones on one earbud and relies on a closed ear canal. The technical problem it addresses is echo suppression within a single ear, not the identification of pathological sound sources in objective tinnitus. This approach neither utilizes the symmetrical difference between pathological and interfering signals in the binaural space nor provides a hardware-triggered recording mechanism for intermittent events.

[0005] The closest prior art to the technical solution of this invention is Canadian patent application CA2952869A1 (whose family includes US10736546B2), which discloses a method and device for continuous in-ear hearing health monitoring on a human being. This system uses a pair of in-ear earplugs, each earplug integrating an external microphone, an internal microphone, and a miniature acoustic transmitter. This solution actively transmits dual excitation tones from the transmitter into the ear canal, and then uses the in-ear microphone to collect the induced distortion product otoacoustic emissions (DPOAE) signal. Based on a normalized least mean square adaptive filter, the external microphone on the same side and the internal microphone on the opposite side are used as reference channels to suppress noise in the collected DPOAE signal, thereby continuously assessing the wearer's hearing health in an industrial noise environment. However, those skilled in the art have always interpreted this solution as merely using the contralateral microphone as a reference channel for environmental noise to improve the signal-to-noise ratio at a single point, a traditional noise reduction paradigm. This solution is essentially a technical approach of active excitation plus adaptive noise reduction, which requires the earplug to occupy and block the external auditory canal, thereby blocking the patient's normal reception of external natural sounds. Furthermore, the use of continuous acquisition throughout the day leads to the accumulation of massive amounts of redundant audio data and hardware power consumption. Moreover, its noise reduction logic cannot effectively separate the weak, intermittent, and irregularly occurring objective tinnitus signals that are truly generated spontaneously by unilateral organic lesions from the massive amount of binaural common-mode sound. For the clinical need of 24-hour long-term unattended monitoring, automatic event interception, and pathological classification of objective tinnitus in everyday natural scenarios, none of the above-mentioned existing technologies provide a feasible solution. Summary of the Invention

[0006] To address the fundamental bottleneck in existing technologies for objective tinnitus detection—which generally rely on a single-point signal-to-noise ratio priority paradigm and fail to identify and engineer the symmetry-breaking of pathological and interference signals in the spatial distribution of both ears—leading to a technical problem where open-type long-range monitoring, hardware-level event filtering, and automatic pathological classification cannot be simultaneously achieved, this invention provides an ear canal sound detection and analysis system. By using two differential components acquired in parallel by binaural probe microphones and binaural reference microphones as hardware triggers, the system directly encodes the pathological prior of unilateral objective tinnitus into the acquisition decision. Under the premise that the open earplug does not close the external auditory canal and does not sacrifice the wearer's perception of natural external sound, it achieves 24-hour long-range unattended capture, precise interception, and pathological classification identification of intermittent objective tinnitus signals from the perspective of event gating driven by spatial symmetry breaking and multi-channel differential self-calibration principles.

[0007] The technical solution of this invention is as follows: an ear canal sound detection and analysis system, comprising a left behind-the-ear main unit and a right behind-the-ear main unit, both of which include a main unit housing; open-back earplugs are connected in pairs to the left behind-the-ear main unit and the right behind-the-ear main unit, each open-back earplug having a probe tube that extends into the same-side external auditory canal without sealing the external auditory canal; probe microphones are arranged in pairs inside the two main unit housings, collecting sound from the same-side ear canal through the same-side probe tube and outputting probe audio signals; reference microphones are arranged in pairs on the outer surface of the two main unit housings, collecting ambient sound from the same-side auricle and outputting reference audio signals; a binaural synchronization module will synchronize the left and right ear canals. The probe audio signal on the right and the reference audio signals on the left and right are time-domain aligned to output a binaural synchronous audio frame. The pathological symmetry defect detection module calculates the ear canal difference degree based on the probe audio signals on the left and right and the environmental difference degree based on the reference audio signals on the left and right. When the ear canal difference degree exceeds a preset pathological threshold and the environmental difference degree is lower than a preset environmental threshold, the trigger recording module starts the truncation and storage of the binaural synchronous audio frame and outputs tinnitus candidate audio segments. The pathological subtyping identification module extracts time-frequency features from the tinnitus candidate audio segments and outputs the sound source side identification and pathological subtyping results.

[0008] Furthermore, the pathological symmetry defect detection module includes a cephalometric registration submodule, an amplitude-frequency compensation submodule, and a binaural transfer function calibration submodule, used to perform temporal domain alignment, external copy amplitude-frequency residual term subtraction, and physiological asymmetry online calibration on the left and right probe audio signals based on the reference audio signal; the pathological typing identification module receives the external physiological reference signal output by the external physiological reference receiving module, and performs mechanism-level typing of the tinnitus candidate audio segment in combination with the rhythm alignment degree; the trigger recording module performs ipsilateral adaptive updates to the preset pathological threshold according to the typing result and the time-frequency feature confidence.

[0009] First, this invention, without sealing the external auditory canal with an open-back earplug, achieves simultaneous output of the ear canal differential degree reflecting the distribution of pathological signals inside the ear canal and the environmental differential degree reflecting the distribution of ambient sound outside the auricle by aligning the paired arrangement of the probe microphone and the reference microphone with the temporal domain of the binaural synchronization module. The mechanism is that objective tinnitus is characterized by unilateral occurrence as a significant pathological feature, while non-pathological interference sources such as ambient sound, breathing, heartbeat, swallowing, and clothing friction have a symmetry between the two ears that can be described by the head shadow effect. The simultaneous existence of the two differential components amplifies the discriminative contrast of the pathological signal relative to the interference signal at the physical layer. This is a fundamental paradigm difference from relying on active stimulation and subjective matching, only performing echo suppression in one earplug, and only using the opposite microphone as a noise reduction reference channel. The improved discriminativeness of the obtained pathological signal cannot be achieved by a single differential, but is the synergistic result of the parallel action of the two differentials.

[0010] Second, the triggering recording module uses the ear canal differential degree exceeding the preset pathological threshold and the environmental differential degree being lower than the preset environmental threshold as a dual-condition triggering condition. Compared with the all-time linear recording mode used in the background technology, its mechanism is that at the hardware level, the joint prior of the purely pathological event of significant asymmetry between the two ears while the external environment remains symmetrical is directly used as the storage start gate. Non-pathological events (such as sudden external sounds and bilateral respiratory noises) cannot meet the dual conditions because they will simultaneously increase the differential degree of both channels, and are therefore automatically filtered out. This significantly reduces the storage capacity, battery power consumption, and subsequent manual screening workload during the 24-hour long-term monitoring process. Its effect cannot be replaced by any single threshold triggering scheme and belongs to the emergent effect of dual threshold joint gating.

[0011] Third, the cephalometric registration submodule within the pathological symmetry defect detection module performs cross-correlation calculations on the reference audio signal as an external binaural copy to generate a time-domain alignment basis. The amplitude-frequency compensation submodule further deducts the external copy amplitude-frequency residual terms of the left and right probe paths. The binaural transfer function calibration submodule only updates the amplitude-frequency transfer function ratio online on the binaural synchronized audio frames where the environmental difference is lower than the preset environmental threshold and the ear canal difference is simultaneously lower than the preset silence threshold. These three components together constitute a closed-loop physiological asymmetry self-calibration link. The mechanism is that the binaural probe paths under open-back earplugs inevitably have amplitude-frequency response asymmetry due to differences in external auditory canal anatomy and wearing depth. Without calibration, this physiological asymmetry will be misjudged as pathological asymmetry. Online calibration only updates the coefficients during the time period determined to be a pure external common-mode segment, which can avoid the calibration itself being contaminated by tinnitus signals. This mechanism-level innovation effectively solves the long-standing problem of physiological asymmetry false triggering in open-back scenarios.

[0012] Fourth, the pathological subtyping identification module, in conjunction with the external physiological reference signal received by the external physiological reference receiving module, performs rhythm alignment calculation on the signal envelope of the tinnitus candidate audio segment and the external physiological reference signal. When the rhythm alignment is greater than the preset vascular synchronization threshold, it outputs vascular tinnitus as the pathological subtyping result. When the spectral peak of the tinnitus candidate audio segment falls within the preset myoclonus frequency band and the rhythm alignment is less than the preset vascular synchronization threshold, it outputs myogenic tinnitus as the pathological subtyping result. The mechanism is to introduce the prior external physiological mechanism of heart rhythm or pulse rate, so that the subtyping decision no longer depends on a single audio feature, but starts from the pathophysiological essence of the pathological source and cardiovascular rhythm synchronization, which significantly improves the robustness of subtyping.

[0013] Fifth, the pathological typing identification module feeds back the pathological typing result and the time-frequency feature confidence level to the trigger recording module. The trigger recording module performs same-side adaptive updates to the preset pathological threshold. The mechanism is that the system as a whole forms a closed-loop optimization loop of detection-typing-threshold feedback: when the same pathological typing result has been repeatedly confirmed on the same side within a unit of time, raising the threshold can suppress redundant triggering. When there is no triggering for a long time or the time-frequency feature confidence level is lower than the preset confidence level lower limit, lowering the threshold can maintain sensitivity. This closed loop makes the trigger specificity and sensitivity for the individual improve synchronously as the system is used for a longer time and the more pathological samples are accumulated, forming a system-level emergent capability that exceeds the sum of the individual effects of each module. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the hardware structure connection between the behind-the-ear host and the open probe earplug of the ear canal sound detection and analysis system of the present invention.

[0015] Figure 2 This is an overall module interaction diagram of the ear canal sound detection and analysis system of the present invention.

[0016] Figure 3 This is a schematic diagram of the internal sub-module structure and data flow of the pathological symmetry defect detection module of the present invention.

[0017] Figure 4 This is a schematic diagram of the rhythm alignment and classification decision process of the pathological classification identification module of the present invention. Detailed Implementation

[0018] The embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. It should be understood that the following embodiments are for illustrative purposes only and are not intended to limit the scope of protection of the present invention.

[0019] See Figure 1 and Figure 2 The ear canal sound detection and analysis system in this embodiment consists of two main functional groups: a hardware acquisition end and a signal processing end. The two are organically coupled through a low-power wireless link to jointly achieve passive capture, event gating, and pathological classification of 24-hour long-range objective tinnitus.

[0020] The hardware acquisition unit consists of the left behind-the-ear (BTE) main unit and the right behind-the-ear (BTE) main unit, both resembling the shape of a behind-the-ear hearing aid. The main unit housing 1 is made of medical-grade ABS or PC / ABS alloy, with the weight of each side controlled between 8g and 12g to ensure comfort during 24-hour continuous wear. The left and right behind-the-ear main units are not connected by any wires; they are independently powered, sample, and buffered. A soft timing relationship is established only through the wireless synchronization time base of the binaural synchronization module. This design avoids the cross-cranial connection harness required by traditional wired binaural interconnection devices, significantly reducing the physical constraint caused by prolonged wear.

[0021] The signal processing unit consists of the binaural synchronization module, the pathological symmetry defect detection module, the trigger recording module, the pathological typing identification module, and the external physiological reference receiving module, which are cascaded and coupled in sequence. The first three are embedded in the embedded processing units inside the main unit housing 1 of the left and right behind-the-ear hosts, forming a real-time event detection pipeline on the edge side. The pathological typing identification module is deployed on the smart terminal (such as a mobile phone or portable computer), and performs offline typing after the tinnitus candidate audio segment is forwarded by the low-power Bluetooth module. The external physiological reference receiving module can be built into the left and right behind-the-ear hosts (when a photoplethysmography sensor is used to obtain the pulse signal), or it can be independently on the smart terminal to obtain the ECG signal by pairing with the ECG patch via low-power Bluetooth. The data flow between modules strictly follows a closed-loop link: probe audio signal / reference audio signal → binaural synchronous audio frame → ear canal differential degree / environmental differential degree → tinnitus candidate audio segment → time-frequency features and signal envelope → pathological classification results and time-frequency feature confidence degree → preset pathological threshold feedback update amount, ensuring no floating input.

[0022] The system as a whole constitutes a deeply coupled closed-loop collaborative architecture: the front input source of the pathological symmetry defect detection module is the binaural synchronization audio frame output by the binaural synchronization module; its rear output, the ear canal difference degree and the environmental difference degree, are used as the decision input of the trigger recording module; the output of the trigger recording module, the tinnitus candidate audio segment, becomes the processing object of the pathological subtyping identification module; the rear output of the pathological subtyping identification module, the pathological subtyping result and the time-frequency feature confidence degree, are then transmitted back to the trigger recording module to adjust the preset pathological threshold; the binaural transfer function calibration submodule inside the pathological symmetry defect detection module updates its own amplitude-frequency coefficient based on the judgment result of the current binaural synchronization audio frame by the trigger recording module, thus the front and rear modules are coupled together to form a closed-loop system with both feedforward and feedback.

[0023] See Figure 1The left-side behind-the-ear main unit's housing 1 houses the probe microphone 101 at the upper part and the reference microphone 102 at the lower part. The right-side behind-the-ear main unit adopts a mirror-symmetrical hardware layout. The open-back earplug 2 is made of medical-grade silicone and is designed with multiple sets of longitudinal ventilation grooves. These grooves maintain an air communication gap with the external auditory canal wall, allowing external sound waves to freely reach the eardrum even when the open-back earplug 2 is fully inserted into the external auditory canal, thus preserving the wearer's normal perception of natural sounds. The probe tube 3 is made of polytetrafluoroethylene or medical-grade polyurethane, with a diameter of approximately 1.5 mm and a wall thickness of approximately 0.2 mm. The tube's interior is an air channel, with one end connected to the microphone array at the front of the open-back earplug 2, and the other end reaching the pickup surface of the probe microphone 101 through the upper acoustic coupling opening of the main unit housing 1. The probe tube 3 is designed to be 35mm to 45mm long, so that after it is inserted into the external auditory canal, the sound receiving hole is located in a high sound pressure area about 5mm to 10mm away from the tympanic membrane.

[0024] The probe microphone 101 is preferably a low-noise silicon-based microelectromechanical microphone with a sensitivity of not less than -38dBV / Pa, a signal-to-noise ratio of not less than 65dB, and a frequency response range covering 20Hz to 10kHz. The reference microphone 102 uses the same model or equivalent signal-to-noise ratio as the probe microphone 101 to ensure consistent noise floor characteristics between the two audio signals. The outer surface of the main housing 1 is provided with a thin-hole array for wind and droplet protection at the corresponding position of the reference microphone 102, which ensures an acoustic path for ambient sound while preventing environmental disturbances from directly impacting the pickup surface.

[0025] The open-back earplug 2 is available in three sizes: S, M, and L, which correspond to the effective diameter of the external auditory canal of approximately 6mm, 7.5mm, and 9mm, respectively. None of the earplug sizes close the external auditory canal.

[0026] The main unit housing 1 integrates a circuit board with an embedded low-power processor at its core. The processor's operating frequency can be dynamically adjusted according to the workload: it drops to 32MHz in non-triggered state to reduce power consumption, and instantly increases to 160MHz after an event is triggered to ensure high-definition sampling and storage. On-chip storage includes a circular RAM buffer for caching the binaural synchronized audio frames and NAND flash memory for storing candidate tinnitus audio segments. The flash memory capacity is 2GB to 8GB, sufficient to support continuous 24-hour event-triggered storage. The low-power Bluetooth module complies with Bluetooth Low Energy 5.0 or higher specifications, employing different low-power strategies in the broadcast-connection-transmission stages. The left and right earbud main units each have built-in batteries, preferably cylindrical lithium polymer cells with a capacity of not less than 100mA·h. The built-in batteries support continuous operation for more than 24 hours when fully charged.

[0027] The core function of the binaural synchronization module is to establish soft timing alignment between the physically separated left and right back-of-ear hosts, so that the probe audio signals and the reference audio signals collected from both sides can be processed in pairs by the pathological symmetry defect detection module under a unified time base.

[0028] The binaural synchronization module adopts a master-slave timing broadcast mechanism: one of the left and right behind-the-ear hosts is designated as the master time base source. The master time base source sends timing anchor frames to the other side through the low-power Bluetooth module at a preset broadcast period (e.g., 100ms). After receiving the timing anchor frames, the other side performs Kalman smoothing update on its local sampling clock to stabilize the instantaneous deviation of the sampling clocks on both sides within ±25μs.

[0029] At the sampling end, both the left and right behind-the-ear (BTE) hosts synchronously sample their respective probe audio signals and reference audio signals at a sampling rate of 16kHz or 48kHz. Each sample of 4096 points forms a raw material block for a binaural synchronized audio frame. Material blocks with the same sequence number on both sides are identified as binaural pairs at the same time. The binaural synchronization module applies a uniform framing window function (such as a Hann window) to the paired material blocks, outputting binaural synchronized audio frames with a frame length of 256ms and an overlap rate of 50% between adjacent frames. The internal structure of each binaural synchronized audio frame includes four time-aligned discrete-time domain audio sequences: a left probe subframe, a right probe subframe, a left reference subframe, and a right reference subframe.

[0030] The binaural synchronization module is also responsible for frame loss compensation: when the low-power Bluetooth module experiences instantaneous link fluctuations that cause a delay in the arrival of a material block on one side, the binaural synchronization module performs linear prediction interpolation on the missing subframes. The binaural synchronization audio frames with an interpolation length exceeding 50ms are marked as unreliable frames and do not enter the differential calculation pipeline of the pathological symmetry defect detection module.

[0031] The pathological symmetry defect detection module is the core processing module of this invention, and its structure is described below. Figure 3 Internally, it includes three functional units: the head shadow registration submodule, the amplitude-frequency compensation submodule, and the binaural transfer function calibration submodule. These three units form a time-series nested coupling relationship: First, the head shadow registration submodule calculates the instantaneous delay of the head shadow effect in the current binaural synchronized audio frame based on the reference audio signal, and performs time-domain alignment on the left and right probe audio signals accordingly; then, the amplitude-frequency compensation submodule subtracts the external replica amplitude-frequency residual terms of the left and right probe paths in the frequency domain; the binaural transfer function calibration submodule determines whether to update the amplitude-frequency transfer function ratio with the reference amplitude-frequency ratio of the current frame based on the feedback from the trigger recording module on the current frame.

[0032] Objective tinnitus (especially vascular and myogenic tinnitus) is characterized by significant unilateral episodes. When external ambient sound acts on the left and right auricles via the head shadow effect, a time-domain lag, mainly determined by the sound source location and slowly varying over time, is formed between the reference audio signals acquired by the left and right reference microphones 102. If this lag is not compensated for, directly differentiating the left and right probe audio signals will treat the external binaural copies themselves as pathological abnormalities, resulting in numerous false triggers.

[0033] The head shadow registration submodule performs a short-time cross-correlation operation on the left and right reference audio signals for each of the binaural synchronized audio frames, and takes the time-domain lag corresponding to the maximum value of the cross-correlation function as the instantaneous delay of the head shadow effect. This delay typically varies between -0.6ms and 0.6ms (corresponding to the geometric relationship between a distance of approximately 20cm between the ears and the speed of sound of 340m / s). The head shadow registration submodule then performs a time-domain translation of the left probe audio signal according to the instantaneous delay of the head shadow effect, so that it is time-domain aligned with the right probe audio signal in the external common-mode sound portion. After alignment, the pathological symmetry defect detection module calculates the ear canal differential degree.

[0034] Since cross-correlation is a well-known basic algorithm in digital signal processing, the formula details will not be elaborated here. However, it should be noted that the head shadow registration submodule uses parabolic interpolation for the cross-correlation peak to achieve sub-sampling point accuracy, thereby suppressing the delay estimation error to less than 1 / 10 of the sampling period.

[0035] Even though the temporal domain has been aligned by the cephalometric registration submodule, the external ear copies acquired by the left and right probe microphones 101 will still exhibit asymmetry in amplitude-frequency response due to anatomical differences between the left and right external auditory canals, differences in the insertion depth of the probe tubing 3, and individual differences in the probe microphones 101. This physiological asymmetry, once mixed with pathological asymmetry, will severely interfere with the physical meaning of the ear canal differential. The task of the amplitude-frequency compensation submodule is to quantitatively estimate and subtract this asymmetric component in the frequency domain.

[0036] Let the short-time Fourier transform result of the probe audio signal on the left side on the binaural synchronized audio frame be... The short-time Fourier transform result corresponding to the probe audio signal on the right is: Similarly, the short-time Fourier transform results corresponding to the reference audio signals mentioned on the left and right are respectively denoted as... and The amplitude-frequency compensation submodule is based on the amplitude-frequency transfer function ratio provided by the binaural transfer function calibration submodule. For each frequency point Calculate the amplitude-frequency residual term of the external replica: , in: The external replica amplitude-frequency residual term is a scalar with a range of values. The unit is V (normalized sound pressure level), calculated by this formula, representing the current binaural synchronized audio frame. Frame number The residual energy of pure external common-mode sound at a frequency point that cannot be resolved by amplitude-frequency ratio between the left and right reference audio signals reflects the residual asymmetry of the left and right reference microphone 102 paths. The reference audio signal described on the left is in the... Frame number The short-time Fourier transform coefficients at a frequency point are complex scalars with a range of values. The unit is V, which is obtained by performing a discrete Fourier transform on the reference audio signal on the left by framing and windowing, and represents the frequency domain component of the reference path on the left. The reference audio signal described on the right is in the... Frame number The short-time Fourier transform coefficients at frequency points, their meanings and... Mirror symmetry; The amplitude-frequency transfer function ratio is a complex scalar function with a range of values. (Physically feasible region), dimensionless, obtained by online updating of the binaural transfer function calibration submodule, characterizing the frequency point of the reference microphone 102 path on the left relative to the reference microphone 102 path on the right. Amplitude-frequency and phase difference correction factor at the location; For frequency index, is an integer scalar, and its value range is... ,in The number of points in the short-time Fourier transform (in this embodiment, we take...) (), dimensionless, determined by the number of transformation points, representing frequency domain coordinates; Frame index, integer scalar, value range Dimensionless, determined by increasing frame number, representing time coordinates; The modulo operator for complex numbers. (Left side of the equation) The unit is V; the right side of the equation The unit is V. The unit is (dimensionless) V) V, the difference between the two is still V, and the modulus is still V; the units on both sides are V, and the dimensions are consistent.

[0037] Based on the external replica amplitude-frequency residual term, the amplitude-frequency compensation submodule further calculates the pathological net difference scale: ,in: The pathological net difference scale is a non-negative real scalar with a range of values. The unit is V, which is calculated by this formula and represents the pure pathological binaural asymmetric energy retained by the current binaural synchronized audio frame after deducting the external copy amplitude-frequency residual term. The probe audio signal described on the left is in the... Frame number The short-time Fourier transform coefficients at the frequency point are complex scalars with units of V; their meanings are otherwise consistent with... correspond; The short-time Fourier transform coefficients corresponding to the probe audio signal described on the right are, with the following meanings: Mirror symmetry; The definition is the same as the formula mentioned above; The definition is the same as the formula mentioned above; Here, represents the frequency weighting coefficients, and represents a real scalar function with a range of values. Dimensionless, determined a priori by the pathological frequency band (within the preset myoclonus frequency band and vascular pulsation frequency band). Other frequency bands are linearly decayed to 0.3), representing the weight of the contribution of different frequency points to the net difference in the pathology. Here, is the residual term suppression coefficient, and is a real scalar parameter with a range of values. Dimensionless, obtained through offline calibration and adjustable during use. Values ​​that are too small will result in insufficient subtraction of physiological asymmetry, while values ​​that are too large will subtract true pathological signals. Optimal... ; and These are the low-frequency and high-frequency integral boundary indices, both integer scalars, with values ​​covering the frequency points from 20Hz to 4kHz. They are dimensionless and determined by the sampling rate and the number of points in the short-time Fourier transform, representing the pathologically suspicious frequency bands. The operator that takes the maximum of the two values; This represents the summation operator. The left side of the equation... The unit is V; in the summation term on the right side of the equation The unit is V. The unit is (dimensionless) V) V, the difference between the two, and the non-negative value is still V. Dimensionless multiplication still results in V, and summation does not change the dimensions; both sides have units of V, so the dimensions are consistent.

[0038] The amplitude-frequency compensation submodule uses the pathological net difference scale. The workload output of the ear canal differential degree is provided to the trigger recording module for comparison with the preset pathological threshold.

[0039] The binaural transfer function calibration submodule is responsible for online estimation and maintenance of the aforementioned... Its core innovation lies in employing an event-gated selective update mechanism, which only updates the binaural synchronized audio frame within a brief window when the current frame is determined to be a pure external common-mode signal with no unilateral abnormality. Incremental updates are performed to avoid the calibration process being contaminated by the objective tinnitus signal itself.

[0040] Specifically, the binaural transfer function calibration submodule performs the following processing on each binaural synchronized audio frame. First, a window gating indicator function is defined. : , in: The calibration window gating indicator function is a binary scalar with a value of [value missing]. , dimensionless, is determined by this formula. When its value is 1, it means that the current binaural synchronized audio frame can be used for calibration update. When its value is 0, it means that the frame is excluded from calibration update. The environmental differential degree is a non-negative real scalar with the unit V. It is obtained by the pathological symmetry defect detection module by performing frequency domain difference and wide-band summation on the left and right reference audio signals, and characterizes the asymmetric energy of the external sound field between the left and right reference microphones 102. The original ear canal differential value before deduction, is a non-negative real scalar value in V, derived from... The calculation yields the total asymmetric energy of the binaural probes in the current frame without deducting the amplitude-frequency residual term of the external replica. The preset environmental threshold is a real scalar parameter with a value range of approximately [value missing]. (Normalized sound pressure unit), the unit is V, which is determined by offline calibration or user-customized configuration, and is used to determine whether the external sound field maintains binaural symmetry; The preset silence threshold is a real scalar parameter with a value range of approximately [value missing]. The unit is V, determined by offline calibration, and is used to determine whether the ear canal is in a silent state without pathological attacks in the current frame. Both sides of the comparison operation are V in dimension, and the judgment result is a dimensionless binary indicator function with consistent dimensions.

[0041] Subsequently, the binaural transfer function calibration submodule is based on the indicator function. Perform a moving average update: , in: The ratio of the amplitude-frequency transfer function used in the next binaural synchronized audio frame is a complex scalar function with a range of values. Dimensionless, obtained by updating this formula, and its meaning is the same as described above. ; The ratio of the amplitude-frequency transfer function used in the current frame; The moving average learning rate is denoted as , where is a real scalar parameter with a range of values. Dimensionless, determined by a trade-off between convergence speed and noise resistance, with optimal value. If the value is too large, the calibration coefficient will be too sensitive to instantaneous disturbances; if it is too small, the ability to track changes in the wearing position will be insufficient. and The definition is the same as the formula mentioned above; To avoid dividing by zero, small positive constants are used as real scalars with a range of values. The unit is V, which is determined by the numerical stability requirements and characterizes the numerical protection for extremely low amplitude reference signals. The definition is the same as in the aforementioned formula. Dimensionless; The unit is V / V Dimensionless; Dimensionless and Dimensionless products remain dimensionless. Multiplying two dimensionless ratios results in a dimensionless product; adding two items results in a dimensionless product; both sides are dimensionless, meaning they have the same dimensions.

[0042] The initialization strategy of the binaural transfer function calibration submodule is as follows: within the first 30 seconds after system power-on, it is forced to... To quickly establish initial calibration coefficients, this stage requires the wearer to be in a normal daily life scenario without tinnitus attacks; after 30 seconds, the binaural transfer function calibration submodule automatically switches to event gating mode, and during long-term operation, the coefficients are updated only on the selected binaural synchronized audio frames.

[0043] The timing logic of the binaural transfer function calibration submodule clearly demonstrates the three stages of perception, gating, and updating: the perception stage receives four subframes from the binaural synchronization module; the gating stage receives the judgment result of the current frame from the trigger recording module, or is directly determined by the pathological symmetry defect detection module. The update phase executes a moving average of the coefficients.

[0044] The trigger recording module is a hardware gating mechanism for tinnitus events to be stored in long-term storage. Its core decision logic is a dual-threshold joint trigger: triggering only when the ear canal differential degree... (Right now The pathological net difference score obtained after deduction by the amplitude-frequency compensation submodule exceeds the preset pathological threshold. And at the same time, the environmental differentiation degree Below the preset environmental threshold Only when the time is right, the capture and storage of the current binaural synchronized audio frame and the audio data 2 seconds before and after it is initiated, and the captured audio segment is marked as the tinnitus candidate audio segment.

[0045] The preset pathological threshold These are not fixed constants, but rather side-by-side parameters maintained by the trigger recording module. During the initialization phase, The values ​​on both sides are given by the offline calibration, with the preferred initial value being [value to be filled in]. (Normalized sound pressure unit). During system operation, the trigger recording module receives the pathological classification result and the time-frequency feature confidence level given by the pathological classification identification module for each tinnitus candidate audio segment. The preset pathological thresholds are updated ipsilaterally according to the following update rules: ,in: For the preset pathological threshold at time against The updated value on the side (left or right) is a real scalar with a range of values. (Preferred) , The unit is V, calculated by this formula, representing the current target... Hardware trigger sensitivity on the side; For the preset pathological threshold at time The current value; For lateral indexes, for symbolic scalars, the value is... , dimensionless, determined by the side of the sound source, indicating that the threshold only applies to the same side; For the current moment, For the threshold update cycle, the preferred method is... s; The unit time window length is a real scalar, and the preferred value is... min, in seconds, represents the observation window for statistically repeated events; For at any time The length forward is Within the time window The number of tinnitus candidate audio segments that have been output with the same pathological classification result is a non-negative integer scalar, which is accumulated from the output of the pathological classification recognition module. Within the same time window The total number of times the event occurs is a non-negative integer scalar. Within the same time window The arithmetic mean of the confidence scores of the time-frequency features mentioned above is a real scalar with a range of values. Dimensionless; Let be the repetition threshold, which is a non-negative integer scalar, preferably... This represents the lower limit of the number of triggers that are considered redundant and repetitive. The preset confidence lower limit is a real scalar with a value range of [value range missing]. Preferred Dimensionless, determined by prior labeling of the classifier; and These are the upward adjustment coefficient and the downward adjustment coefficient, both of which are positive real scalar parameters with a range of values. , Preferred , , dimensionless, determined by system stability requirements. Left side of the equation. The unit is V; the right side of the equation The unit is V, and the product term thereafter... Dimensionless ( Dimensionless, degree difference is dimensionless. It is dimensionless, and its product with V is still V; both sides are V, so they have the same dimension.

[0046] The semantics of the update rule are as follows: when the number of triggers for the same pathological classification result on the same side within a unit time exceeds the repetition threshold, the preset pathological threshold on that side is increased to suppress subsequent redundant triggers and reduce storage consumption; when the same side does not trigger even once within a unit time or the average confidence of the time-frequency feature is lower than the preset confidence lower limit, the preset pathological threshold on that side is decreased to maintain sensitivity and avoid missing weak attacks due to an excessively high initial threshold. The trigger recording module also... Apply hard boundary and Limit the amplitude to prevent long-term feedback drift to unreasonable values.

[0047] The closed-loop feedback process of the trigger recording module demonstrates a complete control loop from current frame trigger decision - tinnitus candidate audio segment extraction - pathological subtyping identification - confidence feedback - ipsilateral threshold update.

[0048] After being captured, the candidate tinnitus audio segments are sent to the NAND flash memory inside the host housing 1 and stored in the low-power audio encoding format. The low-power audio encoding format preferably uses an Opus or LC3 encoder, employing variable bit rate encoding with a bit rate not exceeding 64kbps. While ensuring the integrity and fidelity of the tinnitus signal's time-frequency information, the average size of each segment can be controlled within 32kB, sufficient to support local caching of hundreds of events accumulated within a single 24-hour period. The trigger recording module includes the low-power Bluetooth module, which forwards the cached candidate tinnitus audio segments to the smart terminal in real-time transmission mode or batch transmission mode. The smart terminal then runs the pathological subtyping identification module. The real-time transmission mode is suitable for clinical or inpatient scenarios where the smart terminal is always within communication range; the batch transmission mode is suitable for scenarios where the smart terminal is offline or the patient wears it in a community environment. The low-power Bluetooth module forwards the accumulated candidate tinnitus audio segments all at once when the smart terminal reappears within communication range.

[0049] The pathological subtyping identification module runs on the smart terminal, performing time-frequency feature extraction and pathological subtyping decision on the tinnitus candidate audio segments forwarded by the low-power Bluetooth module, and outputting the sound source localization and the pathological subtyping result. See also Figure 4 The pathological typing identification module contains, in sequence, a preprocessing sub-module, a time-frequency feature extraction sub-module, a rhythm alignment sub-module, and a typing decision sub-module.

[0050] The preprocessing subunit performs bandpass filtering, amplitude normalization, and DC removal on the candidate tinnitus audio segments. The bandpass range is preferably 20Hz to 2kHz, based on the preset combination of the myoclonus frequency band and the vascular pulsation frequency band. The time-frequency feature extraction subunit performs short-time Fourier transform on the preprocessed candidate tinnitus audio segments and extracts time-frequency features, including: the energy distribution ratio of the segment in the left and right pathways (used to determine the sound source side), the position of the spectral peak, the centroid of the power spectrum, the spectral entropy, and the zero-crossing rate in the time domain. These features are used by the classification decision subunit for mechanistic-level decision-making.

[0051] The pathological typing identification module receives the external physiological reference signal output by the external physiological reference receiving module, wherein the external physiological reference signal is an electrocardiogram signal or a pulse signal. The rhythm alignment sub-unit extracts the signal envelope from the tinnitus candidate audio segment. Specifically, it takes the Hilbert transform of the tinnitus candidate audio segment to obtain the analytic signal, then takes the modulus to obtain the instantaneous amplitude envelope, and performs second-order low-pass filtering to smooth the instantaneous amplitude envelope. The rhythm alignment sub-unit then performs the rhythm alignment degree calculation between the signal envelope and the external physiological reference signal. , in: Let be the rhythm alignment degree, be a real scalar, and take a value range of . Dimensionless, calculated by this formula, characterizing the signal envelope. With the external physiological reference signal In optimal relative delay The normalized cross-correlation peak value is closer to 1, which means that the rhythms of the two are more synchronized. Let be the candidate relative time delay between two signals, which is an integer scalar with a value range of . The unit is sampling points, which are obtained by enumeration during the search process; The maximum search delay is a positive integer scalar, and the preferred value is... (Corresponding to a physiological rhythm propagation delay window of approximately 250ms, converted at a 16kHz sampling rate), the unit is sampling points; The optimal relative time delay for maximizing the molecule is an integer scalar. The signal envelope is in the first... The values ​​of each sampling point are non-negative real scalars, and their ranges are... The unit is V (normalized sound pressure amplitude unit), obtained by Hilbert transform modulus taking and low-pass filtering; For the external physiological reference signal in the first The value of each sampling point is a real scalar, and the range of values ​​is determined by the selected physiological signal channel. The unit is V (electrocardiogram signal) or au (normalized unit of pulse signal), which is provided by the external physiological reference receiving module. and They are respectively and The arithmetic mean within the observation interval; To observe the total number of samples, which is a positive integer scalar, the preferred value is... The number of samples corresponding to 3s to 5s is dimensionless and determined by the observation window length and sampling rate; The sample index is an integer scalar. Indicates about The maximum value operator. The numerator is... Summation, unit: V V or V au, collectively referred to as V u (where u is any unit of amplitude); the denominator is the product of the square roots of the two variances, also in V. u; Divide the numerator by the denominator to obtain a dimensionless ratio; take The operation does not change the dimensions. Both sides are dimensionless and have the same dimensions.

[0052] The subtyping decision subunit determines the pathological subtyping result based on the rhythm alignment and the time-frequency characteristics. When When tinnitus is diagnosed as vascular tinnitus, the mechanism lies in the fact that the tinnitus signal generated by vascular pulsation is necessarily synchronized with the cardiac cycle. In this case, the signal envelope should maintain a strong correlation with the external physiological reference signal. The preset vascular synchronization threshold has a range of values. Preferred When the spectral peak of the tinnitus candidate audio segment falls within the preset myoclonic frequency band (preferably 30Hz to 500Hz) and When tinnitus is diagnosed as myogenic tinnitus, the mechanism lies in the clonic sound produced by palatine myoclonus or stapedius muscle spasm, which has a significant rhythmic energy accumulation in the range of 30Hz to 500Hz. However, its rhythm originates from the autonomic clonic contraction of local muscle groups rather than cardiovascular rhythms, and therefore is not synchronized with the external physiological reference signal. For the tinnitus candidate audio segments that do not belong to the above two categories, the classification decision subunit marks the pathological classification results as other pathological types or pending review.

[0053] Simultaneously, the classification decision subunit calculates the confidence level of the time-frequency features. Its value is derived from the inverse normalized form of the time-frequency feature clustering distance, and its range is [missing value]. The time-frequency feature confidence level reflects the degree of certainty that the current tinnitus candidate audio segment is correctly classified by the pathological subtyping identification module. This value, along with the pathological subtyping result, is transmitted back to the triggering and recognition module through the low-power Bluetooth module for it to perform the aforementioned ipsilateral adaptive update.

[0054] The external physiological reference receiving module can be built into either the left or right behind-the-ear host, specifically implemented as a photoplethysmography (PPG) sensor on the concha side, using near-infrared LEDs and photodiodes to collect PPG data from subcutaneous blood vessels in the auricle; alternatively, it can be implemented as an external ECG patch paired with the smart terminal, allowing the smart terminal to receive single-lead ECG signals via Bluetooth. The external physiological reference receiving module applies a 50Hz power frequency notch filter and a 0.5Hz high-pass filter to the external physiological reference signal before it is directly used by the pathological typing identification module.

[0055] Example 1: 24-hour monitoring of daily life scenarios Taking a 42-year-old male patient suspected of having vascular tinnitus (who complained of persistent pulsatile tinnitus in his right ear) as an example, this embodiment illustrates the working process of the system in a real-life scenario.

[0056] When a patient first wears the earplugs at the outpatient clinic, a technician selects size M open-back earplug 2 according to the patient's external auditory canal size and assembles them onto the left and right behind-the-ear main units, placing them behind both ears. The external physiological reference receiving module is implemented using the PPG sensor in the left behind-the-ear main unit to collect the pulse signal from the external physiological reference signal. After wearing, a 30-second initialization is completed in a quiet examination room, during which the binaural transfer function calibration submodule establishes its initial state. .

[0057] The patient then returned to their daily life environment, experiencing typical scenarios in sequence, including typing on a keyboard in the office, conversing with colleagues, riding an elevator, traveling in a subway car, having dinner at home, and falling asleep at night. The binaural synchronization module continuously operates at 16kHz sampling and a 256ms frame length; the pathological symmetry defect detection module calculates for each frame. and The trigger recording module determines the trigger according to a dual-threshold joint triggering rule. During keyboard typing in the office, although the keyboard sound has a large instantaneous amplitude, it is a typical sudden external sound, exhibiting a strongly symmetrical distribution on the binaural reference microphone 102. The environmental differential degree... It will briefly rise above the preset environmental threshold. This means that the condition of environmental difference being lower than the environmental threshold is not met, and the triggering is suppressed; during conversations with colleagues, the speech mainly comes from one direction, and the asymmetric reference signal generated by the head shadow effect will also rise synchronously. The triggering is suppressed; it only occurs when the patient's right carotid artery pulsation and right vascular tinnitus occur simultaneously, while the external environment is relatively stable. The right ear probe signal showed a periodic pulsating waveform, which was significantly higher than... , At the same time lower than If both conditions are met simultaneously, the trigger recording module starts the capture of the tinnitus candidate audio segment, each segment lasting 4 seconds (2 seconds before and after the current frame), and saves it to the NAND flash memory using Opus encoding.

[0058] During the daytime, whenever a patient approaches the smart terminal, the low-power Bluetooth module forwards the accumulated tinnitus candidate audio segments in batch transmission mode. The pathological subtyping identification module performs the rhythm alignment calculation on each segment, and detects... Between 0.78 and 0.86, all exceeding [a certain value]. The pathological classification result was determined to be vascular tinnitus, the sound source side was determined to be right, and the confidence level of the time-frequency feature was between 0.82 and 0.93. The trigger recording module accumulated confirmations. Afterwards, since Gradually rise to Subsequently, the number of triggers on the right side per unit time decreased from approximately 40 times per day to approximately 12 times, with a corresponding decrease of approximately 70% in storage usage and Bluetooth transmission power consumption. The left side remained untriggered due to the lack of triggers and the average confidence level of the aforementioned time-frequency characteristics... , Slowly downgraded to To maintain sensitivity to potential left-sided onset, the system, after 24 hours of operation, provides clinicians with an objective diagnostic profile containing approximately 128 audio segments providing evidence of right-sided vascular tinnitus, along with statistical analysis of time-frequency characteristics and rhythm alignment distribution.

[0059] Example 2: Threshold parameter sensitivity comparison This embodiment compares the effects of different parameter configurations on triggering characteristics. Taking a 53-year-old female patient suspected of having myogenic tinnitus (complaining intermittent clicking sounds in the left ear) as an example, two sets of parameters were used to perform 24-hour monitoring.

[0060] Group 1: Initial values ​​of the preset pathological thresholds , , , The second group: the initial value of the preset pathological threshold. The rest is the same as the first group. After monitoring of both groups, the stored results were compared: the first group had a total of 182 triggers, including 148 myogenic tinnitus types and 34 other pathological types; the second group had a total of 71 triggers, including 67 myogenic tinnitus types and 4 other pathological types. It can be seen that the initial threshold of the second group was higher, causing some weaker episodes to not be captured. However, due to the closed-loop feedback mechanism of the trigger recording module, the second group had already captured the triggers by the end of the 24-hour period. Automatically downgraded to The overall coverage converges towards the first group, verifying the self-correction capability of the closed-loop adaptive threshold update mechanism of this invention.

[0061] Further Adjusting the value from 1.0 to 0.6 and repeating the experiment, it was observed that the false trigger rate increased significantly (from about 10% in the first group to about 28%) due to insufficient subtraction of the external replica amplitude-frequency residual term, verifying the necessity of the amplitude-frequency compensation submodule in the entire signal chain.

[0062] The technical value of this invention does not stem from any single innovation, but rather from the deep coupling and closed-loop collaboration among the pathological symmetry defect detection module, the trigger recording module, and the pathological typing identification module. Specifically, the pre-installed pathological symmetry defect detection module, through three sub-modules, ensures the purity of the physical meaning of the ear canal differential degree, providing a solid foundation for the dual-threshold joint decision of the trigger recording module; the trigger recording module, in turn, provides a high-concentration sample for the pathological typing identification module through event gating; and the feedback from the pathological typing identification module, in turn, affects the threshold of the trigger recording module and the calibration of the pathological symmetry defect detection module. The absence of any one of these three components makes it impossible to simultaneously achieve low missed diagnoses, low false alarms, and reliable typing in open, long-range, and natural scenarios.

[0063] The synergistic effect was verified through ablation comparison experiments. Using data from patients with vascular tinnitus in Example 1 as a baseline, the amplitude-frequency compensation submodule, the binaural transfer function calibration submodule, and the closed-loop feedback mechanism were all disabled and then re-run for 24 hours of monitoring. The missed diagnosis rate, false trigger rate, and average confidence level of the time-frequency characteristics were recorded. The complete system had a missed diagnosis rate of approximately 6%, a false trigger rate of approximately 9%, and an average confidence level of 0.87 for the time-frequency characteristics. After disabling the amplitude-frequency compensation submodule, the missed diagnosis rate was approximately 8%, the false trigger rate was approximately 29%, and the average confidence level of the time-frequency characteristics was 0.71. After disabling the binaural transfer function calibration submodule (with calibration coefficients fixed at initial values), the missed diagnosis rate was approximately 14%, the false trigger rate was approximately 21%, and the average confidence level of the time-frequency characteristics was 0.74. After disabling the closed-loop feedback, the missed diagnosis rate was approximately 7%, the false trigger rate was approximately 11%, but the number of redundant repeated triggers per unit time increased by approximately 2.3 times, and storage consumption increased by approximately 60%. These data show that the combined contribution of the three sub-modules and the closed-loop feedback mechanism is far greater than the sum of their individual contributions, significantly demonstrating the deep coupling closed-loop collaborative characteristics of this invention.

[0064] The description of the above embodiments is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made to the above embodiments by those skilled in the art without departing from the spirit of the present invention should be included within the protection scope of the present invention.

Claims

1. An ear canal sound detection and analysis system, characterized in that, include: Both the left and right behind-the-ear main unit include a main unit housing; An open-ended earbud, connected in pairs to the left earbud host and the right earbud host, the open-ended earbud having a probe tube that extends into the same-side external auditory canal without closing the external auditory canal; The probe microphones are arranged in pairs inside the two main unit housings, and collect the sound from the ear canal on the same side through the probe hose on the same side, and output the probe audio signal; Reference microphones are paired and mounted on the outer surfaces of the two main unit housings to collect ambient sound from the same side of the auricle and output a reference audio signal. The binaural synchronization module aligns the audio signals from the left and right probes and the reference audio signals from the left and right probes in the time domain, and outputs binaural synchronized audio frames. The pathological symmetry defect detection module calculates the ear canal difference degree based on the left and right probe audio signals for the binaural synchronous audio frames, and calculates the environmental difference degree based on the left and right reference audio signals; The trigger recording module starts the capture and storage of the binaural synchronous audio frames when the ear canal difference exceeds a preset pathological threshold and the environmental difference is lower than a preset environmental threshold, and outputs tinnitus candidate audio segments. The pathological subtyping identification module extracts time-frequency features from the candidate tinnitus audio segments and outputs the sound source location and pathological subtyping results.

2. The ear canal sound detection and analysis system according to claim 1, characterized in that, The pathological symmetry defect detection module includes a cephalic registration submodule. The cephalic registration submodule uses the reference audio signal as a copy of the external binaural signals, performs cross-correlation on the left and right reference audio signals, and takes the time-domain lag corresponding to the maximum cross-correlation value as the instantaneous delay of the cephalic effect. After the cephalic registration submodule aligns the left and right probe audio signals in the time domain with the instantaneous delay of the cephalic effect, the pathological symmetry defect detection module calculates the ear canal difference degree.

3. The ear canal sound detection and analysis system according to claim 2, characterized in that, The pathological symmetry defect detection module further includes an amplitude-frequency compensation submodule, which estimates the residual amplitude-frequency term of the external copy from the amplitude-frequency ratio of the reference audio signal on the left and the reference audio signal on the right; the pathological symmetry defect detection module subtracts the residual amplitude-frequency term of the external copy from the ear canal difference to obtain the pathological net difference; the trigger recording module compares the pathological net difference with the preset pathological threshold instead of comparing the ear canal difference with the preset pathological threshold.

4. The ear canal sound detection and analysis system according to claim 3, characterized in that, The pathological symmetry defect detection module also includes a binaural transfer function calibration submodule. The binaural transfer function calibration submodule updates the amplitude-frequency transfer function ratio in a sliding average manner only on the binaural synchronized audio frames where the environmental difference is lower than the preset environmental threshold and the ear canal difference is simultaneously lower than the preset silence threshold. The amplitude-frequency compensation submodule calculates the external replica amplitude-frequency residual term using the amplitude-frequency transfer function ratio.

5. The ear canal sound detection and analysis system according to claim 4, characterized in that, It also includes an external physiological reference receiving module, which receives an external physiological reference signal, which is an electrocardiogram signal or a pulse signal; the pathological classification identification module extracts the signal envelope of the tinnitus candidate audio segment and performs a rhythm alignment operation between the signal envelope and the external physiological reference signal; the pathological classification result is vascular tinnitus when the rhythm alignment is greater than a preset vascular synchronization threshold; the pathological classification result is myogenic tinnitus when the spectral peak of the tinnitus candidate audio segment falls within a preset myoclonus frequency band and the rhythm alignment is less than the preset vascular synchronization threshold.

6. The ear canal sound detection and analysis system according to claim 5, characterized in that, The pathological typing identification module feeds back the pathological typing result and the confidence level of the time-frequency feature to the trigger recording module; The trigger recording module performs same-side adaptive updates on the preset pathological threshold based on the accumulated pathological classification results and the time-frequency feature confidence level. The same-side adaptive updates include: increasing the preset pathological threshold corresponding to the sound source side when the number of times the same pathological classification result has been output for the same sound source side within a unit time exceeds a repetition threshold; and decreasing the preset pathological threshold corresponding to the sound source side when no trigger occurs within a unit time or the time-frequency feature confidence level is lower than a preset confidence level lower limit.

7. The ear canal sound detection and analysis system according to claim 1, characterized in that, The open-back earbuds are available in a variety of sizes, each designed to fit different ear canal sizes, and none of the earbud sizes completely close the ear canal.

8. The ear canal sound detection and analysis system according to claim 1, characterized in that, The trigger recording module includes a low-power Bluetooth module, which forwards the tinnitus candidate audio segments to the smart terminal in real-time transmission mode or batch transmission mode, and the smart terminal runs the pathological subtyping identification module.

9. The ear canal sound detection and analysis system according to claim 1, characterized in that, The left and right earbud main units are each equipped with a built-in battery, which can support continuous operation for more than 24 hours when fully charged.

10. The ear canal sound detection and analysis system according to claim 1, characterized in that, The tinnitus candidate audio segments are saved using a low-power audio encoding format, which employs variable bit rate encoding with a bit rate not exceeding 64kbps.

Images